# IMAGING SYNAPSE STRUCTURE AND FUNCTION

EDITED BY: George J. Augustine and Marc Fivaz PUBLISHED IN: Frontiers in Synaptic Neuroscience

#### *Frontiers Copyright Statement*

*© Copyright 2007-2017 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

*The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88945-175-3 DOI 10.3389/978-2-88945-175-3

### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# **IMAGING SYNAPSE STRUCTURE AND FUNCTION**

Topic Editors:

**George J. Augustine,** Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore

**Marc Fivaz,** Duke-NUS Medical School, Singapore

Two complementary super-resolution imaging techniques reveal interactions of single protein molecules with an individual synapse. The scaffold molecule PSD-95 was mapped using live-cell PALM, and the positions of those molecules are plotted color-coded by the density of their neighbors. Trajectories of a probe transmembrane protein were simultaneously measured, using UPAINT technology, and are shown in pink.

Image by Tuo Peter Li and Thomas Blanpied; see article by these authors in this eBook

Development of new imaging technologies in recent years has transformed neuroscience in profound ways. Following on the heels of the revolution based on the Green Fluorescent Protein, refined genetically-encoded fluorescent reporters and genetic targeting strategies now enable optical recording of synaptic transmission in defined neuronal populations at speeds approaching the enviable temporal resolution of electrophysiology. Super-resolution light microscopy permits observation of synapses and their molecular machinery at sub-diffraction resolution. At the ultrastructural level, automated forms of electron microscopy, improvements in specimen fixation methods, and recent efforts to correlate data from light and electron micrographs now make the reconstruction of functional neural circuits a reality. Finally, the use of optogenetic actuators, such as channelrhodopsins, allows precise temporal and spatial manipulation of neuronal activity and is revealing profound insights into the organization of neural circuits and their roles in behavior.

This research topic highlights recent advances in both light and electron microscopy, with a specific focus on approaches that combine innovations from several different fields to obtain novel information about synapse structure and function. We are confident that this collection of articles - three original research papers, six reviews, one methods paper and one perspective article - will enable neuroscientists to achieve the next generation of experiments aimed at cracking the neural code.

**Citation:** Augustine, G. J., Fivaz, M., eds. (2017). Imaging Synapse Structure and Function. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-175-3

# Table of Contents


### **Section 1: Imaging presynaptic signaling**


### **Section 2: Imaging postsynaptic signaling**


Ryuichi Nakajima, Arong Jung, Bong-June Yoon and Bradley J. Baker


### **Section 3: General synapse imaging technologies**

*83 Correlative Light Electron Microscopy: Connecting Synaptic Structure and Function*

Isabell Begemann and Milos Galic


Jung-Hwa Hong and Mikyoung Park

## Editorial: Imaging Synapse Structure and Function

George J. Augustine<sup>1</sup> and Marc Fivaz <sup>2</sup> \*

*<sup>1</sup> Lee Kong Chiang School of Medicine, Nanyang Technological University, Singapore, Singapore, <sup>2</sup> Program in Neuroscience and Behavioral Disorders, Duke-NUS Medical School, Singapore, Singapore*

Keywords: synapse, optical microscopy, electron microscopy, circuits, genetically-encoded biosensors, optogenetics

**Editorial on the Research Topic**

### **Imaging Synapse Structure and Function**

These are the glory days for imaging synapse structure and function. Thanks to recent advances in both optical and electron microscopy, it is now possible to image individual synapses with unprecedented spatial and temporal resolution. The parallel development of a wide range of genetically-encoded synaptic reporters enables all-optical recording of synaptic activity in genetically-defined neuronal populations. These engineering breakthroughs allow neuroscientists to interrogate the brain in ways that were inconceivable just a few years ago. The ability to image synaptic structure and activity in large functional circuits is beginning to yield key insights into how the brain stores, processes, and computes information. This research topic consists of eleven articles (methods, primary research papers, and reviews) that provide an overview of the latest developments in synapse imaging. Rather than attempting an exhaustive list of synaptic reporters and microscopy techniques, our collection emphasizes approaches that merge technical advances from diverse areas to extract a rich palette of novel information from individual synapses.

Watanabe presents a method that combines optogenetics and rapid freezing (Flash-and-Freeze) to visualize the synaptic vesicle (SV) cycle at the ultrastructural level with millisecond resolution Watanabe. This revolutionary approach revealed ultra-fast endocytosis of SVs at central synapses and neuromuscular junctions (Watanabe et al., 2013a,b) and promises to uncover many new kinetic aspects of synapse dynamics. Begemann and Galic review recent efforts to image neuronal preparations with both light and electron microscopy, with a series of hybrid techniques referred to as Correlative Light Electron Microscopy (CLEM). Jackson and Burrone describe the first genetically-encoded fluorescent reporter (sypHy-RGECO) that enables concurrent monitoring of calcium dynamics and SV fusion. sypHy-RGECO will undoubtedly be a powerful means of examining calcium triggering of SV exocytosis at the level of individual presynaptic boutons. Using similar probes, Tang et al. show that the mental disease gene DISC1 (Disrupted-In-Schizophrenia-1) accelerates SV exocytosis by facilitating calcium influx through N-type voltage-gated Ca2<sup>+</sup> channels. Calcium transients at synapses are also shaped by both mobilization and sequestration of calcium by intracellular stores. Kwon et al. report on the latest advances in organelle-specific calcium sensors and review the contribution of the endoplasmic reticulum and mitochondria to calcium dynamics and synaptic transmission/plasticity.

Until recently, one major impediment to imaging of synaptic activity has been our inability to directly measure membrane potential with adequate signal/noise ratio. This is rapidly changing with the recent improvement of a wide range of genetically-encoded voltage indicators (GEVIs) that now are capable of monitoring both single action potentials and even subthreshold synaptic potentials, both in vitro and in vivo Nakajima et al. Three papers describe recent

Edited and reviewed by:

*Per Jesper Sjöström, McGill University, Canada*

\*Correspondence: *Marc Fivaz marc.fivaz@duke-nus.edu.sg*

Received: *19 October 2016* Accepted: *22 November 2016* Published: *15 December 2016*

#### Citation:

*Augustine GJ and Fivaz M (2016) Editorial: Imaging Synapse Structure and Function. Front. Synaptic Neurosci. 8:36. doi: 10.3389/fnsyn.2016.00036* advances on the localization, dynamics and function of postsynaptic receptors and scaffolds. Using a combination of single-molecule tracking (uPAINT) and photoactivated localization microscopy (PALM), Li and Blanpied assess the diffusion properties of membrane proteins within the postsynaptic density (PSD). The same authors recently used localization microscopy to demonstrate the existence of transsynaptic nanocolumns that align the neurotransmitter release machinery to postsynaptic receptors (Tang et al., 2016a). Dosemeci et al. review a series of EM studies that reveal the presence of a dense lamina—the "pallium"—just beneath the core layer of the PSD, and discuss how translocation of signaling proteins and scaffolds in and out of the pallium may shape activity-induced changes in dendritic spines. In keeping with the theme of postsynaptic signaling, Dore et al. discuss evidence for metabotropic functions of NMDARs, based on time-resolved FRET and other imaging approaches. Finally, at the level of synaptic circuits, two reviews describe the use of geneticallyencoded synaptic labels to trace neural circuits in a variety of different model systems, ranging from C. elegans to mammals (Hong and Park; Lee et al.).

### REFERENCES


Overall, we hope that the fine collection of papers contained within this research topic highlights a useful synapse imaging toolkit for the neuroscience community. The next big challenge in brain imaging will be to scale up these synaptic measurements to large ensembles of neurons to comprehend how circuits compute. This will require synaptic reporters that operate in a synaptically-relevant time scale (milliseconds), along with improved genetic targeting strategies, further advances in automated high-speed microscopy, and refined bioinformatics tools for analysis of the resulting large datasets.

## AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

### FUNDING

This work was supported by Singapore Ministry of Education grants MOE2015-T1-001-069 and MOE2015-T2-2-095 to GA and grant MOE2013-T2-1-053 to MF.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Augustine and Fivaz. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Stimulation of Synaptic Vesicle Exocytosis by the Mental Disease Gene DISC1 is Mediated by N-Type Voltage-Gated Calcium Channels

Willcyn Tang1† , Jervis Vermal Thevathasan1† , Qingshu Lin<sup>2</sup> , Kim Buay Lim<sup>1</sup> , Keisuke Kuroda<sup>3</sup> , Kozo Kaibuchi <sup>3</sup> , Marcel Bilger <sup>4</sup> , Tuck Wah Soong<sup>2</sup> and Marc Fivaz 1,2 \*

<sup>1</sup> DUKE-NUS Medical School, Program in Neuroscience and Behavioral Disorders, Singapore, Singapore, <sup>2</sup> Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore, <sup>3</sup> Department of Cell Pharmacology, Nagoya University Graduate School of Medicine, Nagoya, Japan , <sup>4</sup> DUKE-NUS Medical School, Program in Health Services and Systems Research, Singapore, Singapore

Lesions and mutations of the DISC1 (Disrupted-in-schizophrenia-1) gene have been linked to major depression, schizophrenia, bipolar disorder and autism, but the influence of DISC1 on synaptic transmission remains poorly understood. Using two independent genetic approaches—RNAi and a DISC1 KO mouse—we examined the impact of DISC1 on the synaptic vesicle (SV) cycle by population imaging of the synaptic tracer vGpH in hippocampal neurons. DISC1 loss-offunction resulted in a marked decrease in SV exocytic rates during neuronal stimulation and was associated with reduced Ca<sup>2</sup><sup>+</sup> transients at nerve terminals. Impaired SV release was efficiently rescued by elevation of extracellular Ca<sup>2</sup><sup>+</sup>, hinting at a link between DISC1 and voltage-gated Ca<sup>2</sup><sup>+</sup> channels. Accordingly, blockade of N-type Cav2.2 channels mimics and occludes the effect of DISC1 inactivation on SV exocytosis, and overexpression of DISC1 in a heterologous system increases Cav2.2 currents. Collectively, these results show that DISC1-dependent enhancement of SV exocytosis is mediated by Cav2.2 and point to aberrant glutamate release as a probable endophenotype of major psychiatric disorders.

Keywords: DISC1, hippocampus, glutamate, psychiatric disorders, schizophrenia, neurotransmitter, synaptic vesicle release

### INTRODUCTION

Genome-wide association studies and exome sequencing efforts have led to the identification of hundreds of variants associated with psychiatric disorders (International Schizophrenia et al., 2009; Moskvina et al., 2009; Glessner et al., 2010; Fromer et al., 2014; Schizophrenia Working Group of the Psychiatric Genomics, 2014), confirming the complex polygenic nature of these diseases. These genomic data also revealed a significant overlap in genes or gene networks associated with distinct mental illnesses (Cross-Disorder Group of the Psychiatric Genomics et al., 2013), suggesting a common genetic and perhaps circuitry basis for major psychiatric disorders. However, the synaptic and circuitry defects underlying these disorders remain poorly defined, hindering the development of therapeutic solutions.

#### Edited by:

Stéphane Martin, IPMC CNRS UMR7275 - University of Nice Sophia antipolis, France

#### Reviewed by:

Timothy A. Ryan, Weill Medical College of Cornell University, USA Yann Humeau, Centre National de la Recherche Scientifique (CNRS), France

### \*Correspondence:

Marc Fivaz marc.fivaz@duke-nus.edu.sg

†These authors have contributed equally to this work.

Received: 01 February 2016 Accepted: 31 May 2016 Published: 14 June 2016

#### Citation:

Tang W, Thevathasan JV, Lin Q, Lim KB, Kuroda K, Kaibuchi K, Bilger M, Soong TW and Fivaz M (2016) Stimulation of Synaptic Vesicle Exocytosis by the Mental Disease Gene DISC1 is Mediated by N-Type Voltage-Gated Calcium Channels. Front. Synaptic Neurosci. 8:15. doi: 10.3389/fnsyn.2016.00015

DISC1 is the prototypical example of a gene associated with several major psychiatric disorders. It was discovered in a Scottish family at the site of a balanced chromosomal translocation that strongly segregates with major depression, schizophrenia and bipolar disorder (St Clair et al., 1990; Millar et al., 2000, 2001). The high penetrance (∼70%) of this translocation for mental illness supports a causal link between this genetic lesion and major psychiatric conditions (Chubb et al., 2008). DISC1 variants (haplotypes, single nucleotide polymorphisms and copy number variations) have since been independently associated with depression, schizophrenia, bipolar disorders and autism spectrum disorders (Ekelund et al., 2001, 2004; Sachs et al., 2005; Kilpinen et al., 2008). Thus, DISC1 is a major susceptibility factor for mental illness and a relevant genetic entry point to identify core endophenotypes implicated in neuropsychiatric disorders.

The translocation breakpoint in this Scottish family is located in the C-terminal portion of DISC1 and results in overall reduced expression of the full-length transcript and protein (Millar et al., 2005), suggesting that haploinsufficiency is the main mechanism by which this chromosomal alteration confers risk to disease. Alternatively, it has been proposed that a C-terminal truncated form of DISC1 is expressed from the translocated allele and may be pathogenic (Hikida et al., 2007; Pletnikov et al., 2008), although expression of the truncated DISC1 protein in translocation carriers remains to be demonstrated. Consistent with a disease mechanism based on DISC1 loss-of-function, DISC1 expression is also attenuated in human induced pluripotent stem (iPS) cells derived from members of a family with a DISC1 frame-shift mutation that co-segregates with major psychiatric disorders (Wen et al., 2014).

The identification of a large DISC1 interactome consisting of proteins belonging to different ontologic families suggests broad functions of DISC1 in nerve cells. Accordingly, DISC1 has been implicated in multiple aspects of neuronal and brain development, including neurogenesis (Clapcote et al., 2007; Shen et al., 2008; Mao et al., 2009; Singh et al., 2010; Lee et al., 2011), neuronal migration (Kamiya et al., 2005; Duan et al., 2007; Kubo et al., 2010; Steinecke et al., 2012) and maturation (Duan et al., 2007; Shinoda et al., 2007; Niwa et al., 2010). Even though DISC1 interacts with several signaling proteins known to regulate synaptic functions, relatively little is known about functions of DISC1 at the synapse. In particular, the impact of DISC1 on neurotransmitter release remains largely unexplored, despite the fact that aberrant dopamine and glutamate neurotransmission is a probable cause of schizophrenia and other mood disorders (Howes et al., 2015).

Given the preferential expression of DISC1 in the hippocampus and the involvement of this brain structure in cognition and psychiatric disorders (Chubb et al., 2008), we set out to determine the impact of DISC1 on synaptic vesicle (SV) cycling in hippocampal neurons. We used two independent genetic strategies to alter DISC1 expression—RNAi and a DISC1 KO mouse—and imaged SV cycling and Ca2<sup>+</sup> dynamics in large synapse populations. We show that DISC1 elevates synaptic Ca2<sup>+</sup> signals and boosts SV exocytosis at glutamatergic terminals. Our results further indicate that N-type voltage-gated Ca2<sup>+</sup> channels (VGCCs) mediate the stimulatory effect of DISC1 on SV release. These findings identify a central role of DISC1 in neurotransmitter release and provide new insights on the biological basis of synaptic dysfunction in major psychiatric disorders.

### MATERIALS AND METHODS

### DNA, shRNA Constructs, Lentiviruses and Antibodies

The pCAGGs vGlut1-pHluorin (vGpH) and pCAGGs Synaptophysin-GCaMP3 (SyGC3) constructs were gifts from R. Edwards (UCSF; Voglmaier et al., 2006) and S. Voglmaier (UCSF; Li et al., 2011). The pFUGW (Addgene #14883) shRNAexpressing lentiviral vector was modified to express mCherry (pFUmChW). The shRNA targeting sequences are as follows: (1) Scramble 50GGAGCAGACGCTGAATTAC3<sup>0</sup> (Kamiya et al., 2005); (2) DISC1-E 50GGCTACATGAGAAGCACAG3<sup>0</sup> (exon 2; Duan et al., 2007); and (3) DISC1-A 50GGAAGG GCTAGAGATGTTC3<sup>0</sup> (exon 9) designed with Block-it shRNA from Invitrogen. pFUGW scramble shRNA was a gift from A. Sawa (Johns Hopkins). The DISC1 shRNAs constructs were cloned by introducing double-stranded DNA oligos into lentiviral vector pll3.7 (Addgene #11795) using the HpaI and XhoI sites. DNA fragments containing the U6 promoter and shRNAs were then PCR amplified and cloned into pFUmChW using the PacI site. The human DISC1 gene L variant (NCBI Refseq NM018662.2) was PCR amplified from pCMV6-XL5 DISC1-tGFP (Origene) and cloned into pIRES2-DsRed-Express (Clontech) using NheI and SmaI sites. All constructs were sequenced before use. Lentiviral particles expressing pFUmChW shRNAs were prepared as described in Tiscornia et al. (2006). The DNA constructs used for wholecell patch clamp recording are as follows: Cav2.1 (generated in T. W. Soong's lab), Cav2.2 (Addgene #26568), GFP-β2a and α2δ<sup>1</sup> (kindly provided by T. Snutch, UBC). The rabbit polyclonal Abs against Cav2.1 (#ACC-001) and Cav2.2 (#ACC-002) were from Alomone Labs. The polyclonal Ab against the C-terminus of mouse DISC1 was previously described (Kuroda et al., 2011). The polyclonal Ab against human DISC1 (ab59017) and monoclonal Ab against bassoon were from Abcam (ab82958).

### Mouse Lines, Primary Neuron Cultures and Transfection Protocols

DISC1 (∆2–3) mice (C57BL/6JJmsSlc) have been described before Kuroda et al. (2011). Heterozygous DISC1wt/∆2–3 mice were crossed to each other to obtain DISC1wt/wt and DISC1∆2–3/∆2–3 lines from which hippocampal neurons were prepared from E17/E18 embryos, according to published procedures using papain (Worthington) digestion (Huettner and Baughman, 1986). Primary rat hippocampal neurons were prepared from E18 embryos according to Kaech and Banker (2006) and as previously described in Garcia-Alvarez et al. (2015). For live-cell imaging and immunocytochemistry experiments, neurons were grown on poly-L-Lysine coated glass coverslips, on top of a glial feeder according to the Banker Protocol (Kaech and Banker, 2006). For biochemical analysis, cells were seeded on poly-L-Lysine coated 6-well culture dishes. Unless otherwise stated, neurons were cultured for 14–16 days. The vGpH and SyGC3 constructs were electroporated in freshly-dissociated neurons using the Nucleofector kit (Amaxa Biosystems, Lonza). Lentiviral particles expressing shRNAs were added on DIV2, at a MOI (multiplicity of infection) of 1–3. All animal procedures were approved by the SingHealth Institutional Animal Care and Use Committee (IACUC) of Singapore.

### Live-Cell Confocal Imaging and Immunofluorescence

Time-lapse confocal imaging was performed on an inverted Eclipse TE2000-E microscope (Nikon, USA), mounted with a spinning-disk confocal scan head (CSU-10; Yokogawa, Japan), and equipped with a temperature controlled (36.5◦C) stage and an autofocusing system (PFS; Nikon). Images were acquired with an Orca-Flash 4.0 CCD camera (Hamamatsu Photonics, Japan) controlled by MetaMorph 7.8.6 (Molecular Devices, CA, USA) at 0.5 Hz or 20 Hz. Samples were imaged using a 60× (NA 1.4) objective in Tyrode's buffer (150 mM NaCl, 2.5 mM KCl, 1 mM CaCl2, 1 mM MgCl2, 6 mM glucose, 25 mM HEPES, pH 7.4) supplemented with 25 µM 6-cyano-7-nitroquinoxaline-2,3-dione/CNQX (Tocris Bioscience) and 50 µM D,L-2-amino-5-phosphonovaleric acid/AP5 (Tocris Bioscience). Coverslips were mounted in an RC-21BRFS chamber (Warner Instruments, USA) equipped with platinum wire electrodes. Field stimulation was induced by a square pulse stimulator (Grass Technologies, USA) and monitored by an oscilloscope (TDS210, Tektronix, USA). Trains of action potentials (APs) were generated by applying 20 V pulses (1 ms duration) at 10 or 20 Hz. Our typical stimulation paradigm for vGpH measurements involved two consecutive trains of 300 APs at 10 Hz, separated by ∼5 min to allow synapses to recover. vGpH responses between the first and second stimulation were highly reproducible (data not shown). For measuring SV exocytic rates, the second stimulation was preceded (30 s earlier) by the addition of Bafilomycin A1 (Baf; AG scientific, USA) 0.5 µM. To normalize for total expression of vGpH in each individual bouton, 50 mM NH4Cl was added at the end of the time series. For SyGC3 measurements, Ca2<sup>+</sup> signals were normalized by adding 10 µM or 50 µM (for **Figure 4G**) ionomycin (Sigma-Aldrich) at the end of the stimulation protocol. ω-agatoxin TK125 nM (Tocris Bioscience) and ω-conotoxin GVIA 125 nM (Alomone Labs) were used for experiment involving inhibition of P/Q-type Ca2<sup>+</sup> channels and N-type Ca2<sup>+</sup> channels activity, respectively. For immunofluorescence studies, neurons grown on glass coverslips were fixed in 4% paraformaldehyde with 4% sucrose in PBS and permeabilized with 100 ng/ml Digitonin (Sigma-Aldrich). Cells were then incubated with 5% goat serum to block non-specific binding sites and stained with the primary and Alexa Fluor-conjugated secondary antibodies (Life Technologies). Coverslips were then mounted on glass slides and imaged with an inverted laser scanning confocal microscope (LSM710, Zeiss) with a Plan-Apochromat 63× (NA = 1.40) objective.

### Image Analysis

For automated analysis of vGpH responses, we wrote a Matlab script that segments responsive boutons based on the difference between peak vGpH intensity during the first stimulation and baseline intensity prior to stimulation (∆F = Fpeak − Fbaseline). An intensity threshold for ∆F was selected to resolve individual boutons and exclude those with ∆F below 5%. The same threshold was used for all conditions in one independent experiment (i.e., one neuron preparation with control vs. DISC1-depleted conditions). This threshold value was minimally adjusted (less than 10% change) across all independent experiments described in this article. Binarized boutons were then slightly dilated (**Figure 1B**) to ensure that vGpH fluorescence was captured in its totality even after minor lateral movement or change in shape. Time series with minor x–y drifts (originating from the stage) were re-aligned using a script previously described (Thevathasan et al., 2013). Segmented boutons were then size gated, with gating parameters kept constant across all experiments. vGpH fluorescence intensity was then extracted in each segmented bouton across the time series and divided by the signal after NH4Cl to normalize for vGpH expression (**Figure 1C**). This segmentation strategy ensures that the same boutons are analyzed during the two consecutive AP trains. vGpH traces with significant baseline drifts between the first and second stimulation or after Baf application were excluded. A similar segmentation approach was used to analyze SyGC3 Ca2<sup>+</sup> signals. SV exocytic rates were obtained by measuring the slope of the vGpH rise during the second stimulation (in the presence of Baf). The first six time points (during stimulation) were used for linear regression analysis. To measure endocytic rates, the vGpH trace during the first stimulation was subtracted from that obtained during the second stimulation. The resulting trace is a measure of endocytosis (**Figure 1D**). Endocytic rates were measured by linear fitting of six time points chosen after stimulation onset when endocytosis kicks in. Presynaptic localization, abundance and density of Cav2.1 or Cav2.2 were analyzed with a modified version of a Matlab script described previously (Poon et al., 2014). All Matlab scripts are available upon request.

### Immunoblotting

Cultured primary neurons or HEK293T cells were washed with ice-cold PBS and lysed with RIPA buffer (10 mM Tris-HCl pH = 7.2, 150 mM NaCl, 1% TritonX-100, 0.1% SDS, 5 mM EDTA, 0.25% Na-deoxycholate) supplemented with Complete protease inhibitor and PhosphoStop phosphatase inhibitor (Roche). For analysis of hippocampal tissue, the

hippocampi from P10 mice were harvested and homogenized in RIPA buffer using a Dounce tissue homogenizer. Lysates were cleared by centrifugation and boiled in Laemmli sample buffer. Equal amount of total proteins were loaded. Samples were then analyzed by SDS-PAGE, transferred onto nitrocellulose membranes, probed with appropriate primary and HRP-labeled secondary antibodies and revealed by enhanced chemiluminescence.

### Whole-Cell Patch Clamp Recordings of Cav2.1 and Cav2.2 Currents

Cav2.1 or Cav2.2, the auxiliary subunits (GFP-β2a and α2δ1) and DISC1 were transiently transfected in HEK 293 cells using the calcium phosphate method (Huang et al., 2012). Whole-cell patch-clamp recordings were performed within 36–72 h after transfection. The external solution contained (in mM) 10 HEPES, 140 TEA-MeSO<sup>3</sup> and 5 BaCl<sup>2</sup> (pH 7.4, 300–310 mOsm). The glass pipette solution was backfilled with pipette solution (in mM) 10 HEPES, 5 CsCl, 138 Cs-MeSO3, 0.5 EGTA, 1 MgCl2, 2 mg/ml MgATP (pH 7.3, 290–300 mOsm). HEK 293 cells were held at −90 mV using the Axopatch 700B amplifier (Axon Instruments). The series resistance for all recordings was less than 5 MΩ; 70–80% compensation on serial resistance and cell membrane capacitance were applied. A P/4 protocol was used to subtract leakage current. All recordings were obtained with an Axon Digidata 1440A data acquisition system, sampled at 5–50 kHz and low pass-filtered at 1 kHz or 6 kHz. The I-V curves were obtained from 10 mV voltage-steps ranging from −50 to 40 mV and fitted with a modified Boltzmann equation:

$$I = G\_{\text{max}}(E\_{\text{rev}} - V) / \left( 1 + e^{\frac{V\_{1/2\text{act}} - V}{k\_{\text{act}}}} \right)$$

where, I = current density (in pA/pF), Gmax = maximum conductance (in nS/pF), Erev = reversal potential, V = measured potential, V1/2act = midpoint voltage for current activation, and kact = the slope factor.

We used a tail protocol to measure current density; cells were depolarized using 10 mV voltage steps, from −60 to 60 mV. Following depolarization, tail currents were evoked with a 10 ms pulse at −50 mV. The data were fitted with single Boltzmann equation:

$$I = I\_{\rm min} + (I\_{\rm max} - I\_{\rm min}) / \left(1 + e^{\frac{V\_{1/2\rm inact} - V}{k\_{\rm inact}}}\right)$$

where, Imax and I min = maximal and minimal current respectively, V1/2inact = the half-maximal voltage for current inactivation, kinact = slope of inactivation curve.

### Statistics

The experimental design of this study implies that data collected at individual synaptic terminals are clustered according to neuron preparations and imaged fields. Galbraith et al. (2010) demonstrated that such clustering can adversely affect statistical inference when not accounted for. In order to probe for clustering effects (i.e., intra-field correlations), we tested whether bouton responses significantly vary from field to field. For this, we conducted Wald tests of the null hypothesis of no difference in mean outcome across fields within each condition, which revealed (p-value < 0.0001) strong intra-field correlation. To account for these correlations, we performed our statistical analysis in the framework of linear mixed models (Laird and Ware, 1982) with normallydistributed random field effects and preparation fixed effects. In experiments involving one genetically-modified condition and one control group, we tested the null hypothesis of no difference between the mean outcome of the groups via 2-sample t-tests. In experiments involving two geneticallymodified conditions and one control group, we tested the null hypothesis of no difference between the mean outcome of each group and the control group jointly using Wald tests. All tests were performed at the 5% level of statistical significance and carried out using the statistical software Stata version 13.2. Because our statistical approach (linear mixed model) is not a standard practice when analyzing synaptic properties, we compared the p-values obtained with our method with those measured by the more common field averaging approach, where the information of a field is collapsed to a single independent observation by taking the mean of bouton responses. Both methods are valid and gave comparable results (**Supplementary Table T1**), although the field averaging approach is less statistically efficient as it diminishes the information that can be obtained from the data by reducing individual measurements in a field to one observation (Galbraith et al., 2010).

### RESULTS

### DISC1 Loss-of-Function Slows Down SV Cycling

We opted for an imaging approach based on the synaptic tracer vGpH to explore the impact of DISC1 on the SV cycle. vGpH consists of the pH-sensitive GFP variant pHluorin (Miesenbock et al., 1998) fused to the SV-resident glutamate transporter vGlut1. pHluorin, which faces the acidic lumen of SVs, undergoes a ∼20-fold increase in fluorescence intensity when exposed to the neutral pH of the extracellular milieu after membrane fusion. (Sankaranarayanan et al., 2000). Following glutamate discharge and vGpH re-uptake, SVs are rapidly re-acidified and vGpH fluorescence is quenched. This property has made vGpH a valuable tool to monitor both exo- and endocytosis of SVs at single synapses.

Due to the inherent variability in the properties of individual synaptic boutons (Ariel et al., 2013) we measured SV cycling in large synapse populations. For this, we developed an image analysis algorithm that identifies all responding boutons in a given field (**Figures 1A,B**), and imaged 16–18 fields from at six independent rat hippocampal neuron cultures, yielding close to a thousand synaptic boutons for each condition. We employed this algorithm to monitor SV cycling in response to a stimulation paradigm that consists of two consecutive trains of APs (Fernandez-Alfonso and Ryan, 2004; Burrone et al., 2006). The amplitude of the first vGpH transient is governed by the relative rates of SV exo- and endocytosis during stimulation (**Figure 1C**). To separate the contributions of exo- and endocytosis, we blocked SV re-acidification with the vacuolar H<sup>+</sup> ATPase inhibitor Bafilomycin (Baf) during the second stimulation, which allows selective measurement of SV exocytosis (**Figure 1C**). To account for cell-to-cell variation in vGpH expression, we normalized each trace based on the vGpH signal measured after NH4Cl addition (**Figure 1C**). The endocytic component of SV cycling was then computed by subtracting the first vGpH transient from the vGpH response after Baf treatment (**Figure 1D**). Synaptic traces with baseline drifts between the first and second stimulation or after Baf addition were excluded from the analysis (see ''Materials and Methods'' Section). Exocytosis largely dominates at the onset of stimulation, while endocytosis kicks in towards the end of the AP train. Exo- and endocytic rates were finally obtained by linear fitting of the exo- and endo traces (**Figure 1D**). Exocytic rates are about three times higher than endocytic rates under this stimulation protocol. To evaluate the variability of bouton responses across fields and neuron preparations, we compiled exocytic rates from six independent experiments (**Supplementary Figure S1**). Statistical analysis of these responses revealed substantial field to field variation (see ''Materials and Methods'' Section), implying that the information collected from each bouton does not constitute an independent measurement. To compare synaptic responses in different fields and conditions we used a statistical approach that accounts for correlations within fields and variations in neuron preparations (see ''Materials and Methods'' Section and **Supplementary Table T1**).

We then silenced the DISC1 gene using a published shRNA sequence (shRNA-E), targeting exon 2 (Duan et al., 2007) and a new shRNA sequence (shRNA-A) targeting exon 9. These two shRNAs target regions of the DISC1 gene that are fully conserved in rodents and are thus expected to silence rat and mouse DISC1 with the same efficiency. To probe the effect of these shRNAs on DISC1 protein levels, we used an antibody raised against the C-terminus of mouse DISC1, which has previously been validated in a DISC1 KO mouse (Kuroda et al., 2011). In our hands, this antibody poorly reacts with rat DISC1 (not shown). In mouse hippocampal neurons, however, it detects one major protein at ∼100 kDa corresponding to the predicted full length DISC1 (Kuroda et al., 2011) and which is substantially reduced in both shRNA-E and -A transduced cells (**Figure 2A**).

Next, we probed the impact of these shRNAs on the SV cycle in rat hippocampal neurons. We initially chose to work with rat, rather than mouse neurons, because we can harvest substantially more cells from a rat embryonic litter. vGpH signals in shRNA-E and -A expressing neurons display a slower rise and lower amplitude during the first and second stimulation, relative to control cells expressing a scramble shRNA (**Figure 2B**). Analysis of exocytosis shows

substantial synapse-to-synapse variation in all conditions, but reveals a marked decrease in exocytic rates (**Figures 2C,D**) and amplitudes (**Figure 2C**) in both shRNA-E and -A expressing neurons. DISC1 silencing also resulted in slightly slower SV endocytosis, although this effect did not reach statistical significance (**Figures 2E,F**).

Reduced amplitude of the exocytic response prompted us to test whether DISC1 also regulates the size of the total releasable pool. For this, we applied a stimulus strong enough (1200 APs at 10 Hz) to maximally deplete releasable SVs from presynaptic terminals (Ariel and Ryan, 2010). Under this stimulation regime, vGpH responses reached the same plateau in DISC1-silenced and control neurons, albeit with different kinetics, arguing against an effect on the total releasable pool but confirming the stimulatory function of DISC1 in SV exocytosis (**Figure 2G**).

To rule out the possibility of RNAi off-target effects—shRNA-E has recently been suggested to inhibit neuron migration in the developing cortex independently of DISC1 (Tsuboi et al., 2015)—we examined SV cycling in a DISC1 KO mouse that lacks exons 2 and 3 of the DISC1 gene (Kuroda et al., 2011). Homozygous DISC1∆2–3/∆2–3 mice show no detectable levels of the major isoform of DISC1 (**Figure 3A**). Hippocampal neurons derived from DISC1∆2–3/∆2–3 mice display SV cycling defects that are remarkably similar to these observed with RNAi. The rates and amplitudes of exocytic responses were substantially reduced relative to wt cells (**Figures 3B–D**), while endocytic rates were marginally, but not significantly diminished (**Figures 3E,F**). Together, these results show that both genetic ablation and RNAi knockdown of DISC1 selectively disrupts SV exocytosis at glutamatergic synapses with no detectable impact on the total releasable pool.

### DISC1 Stimulates Ca2+ Influx and Regulates Cav2.2-Dependent SV Release

Because SV exocytosis is initiated by Ca2<sup>+</sup> influx through VGCCs, we measured evoked Ca2<sup>+</sup> transients at presynaptic terminals using the SV-targeted Ca2<sup>+</sup> sensor SyGC3 (Li et al., 2011). Ca2<sup>+</sup> signals evoked by 300 APs (10 Hz) or 200 APs (20 Hz) show a rapid rise and partial decay during the stimulus (**Figures 4A,B,D,E**), as observed by others (Li et al., 2011). These Ca2<sup>+</sup> transients were significantly reduced in shRNA-E and -A expressing neurons, under the same stimulation conditions used for vGpH measurements (**Figures 4A,C**), or in response to a higher frequency stimulus (**Figure 4B**). Similarly,

DISC1∆2–3/∆2–3 neurons show lowered Ca2<sup>+</sup> transients compared to wt cells (**Figures 4D–F**). We were concerned that, under such prolonged stimulations, Ca2<sup>+</sup> concentration in terminals might reach levels that partially saturate SyGC3. Ca2<sup>+</sup> signals were therefore also examined in response to shorter stimulations (20 APs, 20 Hz) under conditions well below SyGC3 saturation (Tian et al., 2009; Akerboom et al., 2012). A clear decrease in the amplitude of Ca2<sup>+</sup> transients was also observed in DISC1-silenced neurons under this stimulation regime (**Figure 4G**). Together, these Ca2<sup>+</sup> imaging data hint at a Ca2<sup>+</sup> influx defect in DISC1 inactivated cells, although abnormal clearance of Ca2<sup>+</sup> from the presynaptic terminal could also be involved (see ''Discussion'' Section).

To determine whether reduced Ca2<sup>+</sup> influx is the primary cause of SV cycling defects, we attempted to rescue SV exocytosis in DISC1-silenced neurons by elevating extracellular Ca2<sup>+</sup> concentration. Shifting extracellular Ca2<sup>+</sup> concentration from 2 to 4 mM increases AP-induced Ca2<sup>+</sup> entry (Ariel and Ryan, 2010) and restored the vGpH response (**Figure 4H**), in line with a role of DISC1 in facilitating Ca2<sup>+</sup> influx.

At central synapses, the P/Q-type (Cav2.1) and N-type (Cav2.2) Ca2<sup>+</sup> channels are the main sources of Ca2<sup>+</sup> for initiation of SV exocytosis (Catterall et al., 2013). We used channel-specific toxins to determine the contribution of each subtype to SV release in hippocampal neurons. ω-Conotoxin GVIA (an N-type blocker) reduced SV exocytic rates by 73 ± 5% (**Figures 5A,B**), while ω-Agatoxin TK (a P/Q-type blocker) resulted in a 42 ± 3% decrease in SV exocytic rates (**Figures 5C,D**), in a good agreement with a recent study (Ariel et al., 2013). Note that in these experiments, exocytic rates were approximated by measuring the initial slope of the vGpH response in the absence of Baf (see **Figure 1D**). Importantly, blockade of Cav2.2 almost completely occluded the effect of DISC1 silencing on SV exocytosis (**Figures 5A,B,E**). Blockade of Cav2.1, on the other hand, slightly increased the inhibitory impact of DISC1 knockdown on SV release rates (**Figures 5C–E**). We conclude from these results that DISC1 selectively stimulates Cav2.2-dependent SV release at hippocampal synapses.

We then asked whether DISC1 regulates Cav2.2 activity by controlling its presynaptic localization and/or abundance. Co-localization studies of endogenous Cav2.2 with the active zone marker bassoon revealed extensive presence of Cav2.2 in synaptic boutons, both in control and DISC1-silenced neurons (**Supplementary Figures S2A–C**). We found no significant difference between these two groups in the fraction of boutons

3 exps) and -A (754 boutons, 11 fields, 3 exps) shRNAs, in response to 300 APs, 10 Hz. The DISC1-E and DISC-A groups are significantly different than the scr group (p = 0.0057). (B) Average Ca2<sup>+</sup> transients in neurons expressing scr (861 boutons, 9 fields, 5 exps), DISC1-E (2118 boutons, 7 fields, 3 exps) and -A (1549 boutons, 9 fields, 3 exps) shRNAs, in response to 200 APs, 20 Hz. (C) Cumulative probability of SyGC3 peak intensity from individual boutons corresponding to (A). (D) Average Ca2<sup>+</sup> transients in DISC1 wt/wt (1090 boutons, 11 fields, 3 exps) and DISC1∆2–3/∆2–3 (727 boutons, 9 fields, 3 exps) neurons, in response to 300 APs, 10 Hz. The DISC1 wt/wt and DISC1∆2–3/∆2–3 groups are statistically different (p = 0.0182). (E) Average Ca2<sup>+</sup> transients in DISC1 wt/wt (1365 boutons, 9 fields, 3 exps) and DISC1∆2–3/∆2–3 (1233 boutons, 8 fields, 3 exps) neurons, in response to 200 APs, 20 Hz. (F) Cumulative probability of SyGC3 peak intensity from individual boutons corresponding to (D). (G) Average Ca2<sup>+</sup> transients in neurons expressing scr (290 boutons, 6 fields, 2 exps), and DISC1-E (390 boutons, 6 fields, 2 exps) shRNAs, in response to 20 APs, 20 Hz. The scr and DISC1-E groups are statistically different (p = 0.0036). (H) Average vGpH traces in scr (180 boutons, 5 fields, 2 exps) and DISC1-E (488 boutons, 6 fields, 2 exps) shRNA-expressing neurons during two consecutive trains of APs (300 AP, 10 Hz) in the presence of 2 or 4 mM extracellular Ca2+.

containing Cav2.2, or in the intensity of Cav2.2 staining in presynaptic terminals (**Supplementary Figures S2B,C**). Nor did we find evidence for altered presynaptic localization and intensity of Cav2.1 (**Supplementary Figures S2D–F** or reduced density of presynaptic terminals (**Supplementary Figure S2G**) in DISC1 knockdown neurons. Thus, synaptic targeting and abundance of Cav2 channels does not seem to be regulated by DISC1.

We finally examined the effect of DISC1 on Cav2 currents by whole cell patch-clamp recordings. Co-transfection of recombinant DISC1 (**Figure 6A**) in HEK293T cells together with Cav2.2 and its auxiliary subunits (see ''Materials and Methods'' Section) resulted in a 27% increase in Cav2.2 peak current density (**Figures 6B,C**). Likewise, tail current density of Cav2.2 was substantially elevated in DISC1 overexpressing cells (**Figures 6D,E**). Current density can be

after ω-Agatoxin TK application. (E) Table showing the percentage of inhibition of exocytosis rate by DISC1 knockdown before and after Cav2.2- or Cav2.1 blockade.

affected by a change in channel gating or by the number of channels at the cell surface. To test the first possibility, we examined the voltage-dependent properties of Cav2.2 current activation and saw no difference between control and DISC1-overexpressing cells (**Figure 6F**). This suggests that DISC1 promotes surface delivery and/or stabilize surface expression of Cav2.2. Recording of Cav2.1 currents revealed a similar, voltage-independent, potentiation effect of DISC1 (**Figures 6G–K**). Thus, DISC1 equally augments both Cav2.2 and Cav2.1 currents in this heterologous system, presumably by increasing surface expression of these Ca2<sup>+</sup> channels. These results also imply the presence of an additional layer of regulation in hippocampal neurons that restricts the activity of DISC1 to Cav2.2-dependent SV exocytosis.

## DISCUSSION

We used here two independent gene-targeting approaches together with large-scale imaging of presynaptic function to determine the impact of the mental disease gene DISC1 on the glutamate release machinery. Our results show that DISC1 accelerates SV exocytosis and thus enhances presynaptic performance. This boosting effect is mediated by N-type Ca2<sup>+</sup> channels, establishing the first link between DISC1 and VGCC activity.

FIGURE 6 | DISC1 enhances Cav2.2 and Cav2.1 currents. (A) Western blot showing expression of ectopic (human) DISC1 in HEK293 cells. (B,C) Cav2.2 Current-voltage (I-V) curves for hDISC1-expressing and control cells. (B) Stimulation protocol and individual current responses shown at three different voltages (−30, 0 and 30 mV) (C). Average I-V plots for hDISC1 (peak = 54.2 ± 4.5 pA/pF, 13 cells) and control (peak = 39.2 ± 4.9 pA/pF, 12 cells), p = 0.033. (D–F) Cav2.2 activation curves in response to the tail protocol. (D) Illustration of the tail protocol and individual tail currents measured at −50 mV after three different voltage steps (−20, 0 and 40 mV). (E) Average current density based on tail currents for DISC1 (164.3 ± 7.3 pA/pF, 12 cells) and control (110.5 ± 6.3 pA/pF, 12 cells), <sup>∗</sup>p < 0.001. (F) Normalized activation curve from tail currents showing no significant difference between DISC1 (V50: −6.98 ± 3.43 mV, 12 cells) and control (V50: 0.86 ± 3.62 mV, 12 cells), p = 0.13. (G,H) Cav2.1 Current-voltage (I-V) curves for hDISC1-expressing and control cells. (G) Stimulation protocol and individual current responses shown at three different voltages (−30, 0 and 30 mV). (H) Average I-V plots for hDISC1 (peak = 79.9 ± 8.3 pA/pF, 15 cells) and control (peak = 56.3 ± 7.4 pA/pF, 13 cells), p = 0.046. (I–K) Cav2.1 activation curves in response to the tail protocol. (I) Illustration of the tail protocol and individual tail currents measured at −50 mV after three different voltage steps (−20, 0 and 40 mV). (J) Average current density based on tail currents for DISC1 (156.3 ± 5.7 pA/pF, 25 cells) and control (123.2 ± 7.4 pA/pF, 17 cells), <sup>∗</sup>p = 0.003. (K) Normalized activation curve from tail currents showing no significant difference between DISC1 (V50: −5.39 ± 0.49 mV, 24 cells) and control (V50: −4.41 ± 0.93 mV, 17 cells), p = 0.31.

DISC1 inactivation in hippocampal neurons results in decreased Ca2<sup>+</sup> transients at hippocampal terminals in response to AP firing, consistent with a deficit in Ca2<sup>+</sup> entry. Because these Ca2<sup>+</sup> signals are shaped by both Ca2<sup>+</sup> entry and clearance processes, we cannot exclude a role of DISC1 in regulating Ca2<sup>+</sup> extrusion or buffering. A stimulatory effect of DISC1 on Ca2<sup>+</sup> entry is further supported, however, by its positive impact on Cav2.2 and Cav2.1 currents, the two main VGCCs underlying the initiation of neurotransmission at hippocampal synapses. How DISC1 enhances Cav2 currents is unclear. The lack of an effect on gating suggests that DISC1 promotes surface expression of these Ca2<sup>+</sup> channels, although we cannot exclude a role of DISC1 in regulating single-channel conductance or open probability. Our immunolocalization studies revealed no detectable impact of DISC1 on presynaptic abundance of Cav2 channels. This does not rule out, however, the possibility that DISC1 regulates the number of functional Ca2<sup>+</sup> channels at the surface of the terminal.

Although overexpression of DISC1 enhances both Cav2.1 and Cav2.2 currents in a heterologous system, our findings point to a selective influence of DISC1 on Cav2.2-dependent SV exocytosis. How is this selectivity achieved? It is possible that both Cav2.1 and Cav2.2 compete for DISC1 regulation at presynaptic terminals. The dominant contribution of Cav2.2 to SV release probably results (at least in part) from increased levels of Cav2.2 (relative to Cav2.1) at hippocampal terminals (Ariel et al., 2013). Thus, preferential regulation of Cav2.2 by endogenous levels of DISC1 combined with greater abundance of Cav2.2 may explain why DISC1-dependent SV exocytosis appears to be exclusively mediated by Cav2.2. Specificity could also arise, however, from functions of DISC1 downstream of Ca2<sup>+</sup> entry. For example, DISC1 could be involved in positioning Cav2.2 in close proximity to the SV release machinery at the active zone, a process that dictates speed and fidelity of neurotransmission. While the role of Ca2<sup>+</sup> entry in SV exocytosis is undisputed, its impact on SV endocytosis remains somewhat controversial. Recent evidence suggests, however, that Ca2<sup>+</sup> couples rates of SV exo- and endocytosis and optimizes endocytic rates during AP bursts (Armbruster et al., 2013). We observed reduced endocytic rates in both DISC1-silenced and DISC1∆2–3/∆2–3 neurons. Although this effect was small and did not reach statistical significance, our results suggest that the influence of DISC1 on Cav2.2 impacts both exo- and endocytosis of SVs at hippocampal terminals.

Our findings extend on two recent studies implicating DISC1 in presynaptic function. In glutamatergic neurons differentiated from human iPS cells and derived from members of a family with a frameshift mutation in DISC1, deficits in SV release were observed after KCl-induced depolarization (Wen et al., 2014). In a separate report, RNAi knockdown of DISC1 in layer 2/3 neocortical neurons increases pairedpulse facilitation and appears to reduce probability of glutamate release (Maher and Loturco, 2012). Collectively, these results clearly identify DISC1 as a positive modulator of glutamate release.

The efficacy of neurotransmitter release determines not only the strength of synaptic excitation, but also dictates various forms of short-term plasticity (Abbott and Regehr, 2004), suggesting broad functions of DISC1 in synaptic computation and neural circuit performance. Although we have not directly measured release probability (Pr), the impact of DISC1 on Ca2<sup>+</sup> entry and SV cycling suggests a positive influence on P<sup>r</sup> (see Maher and Loturco, 2012). Because synaptic facilitation and depression depend largely on the initial P<sup>r</sup> (high P<sup>r</sup> favors depression while low P<sup>r</sup> favors facilitation), our results predict altered shortterm plasticity in DISC1-deficient neural circuits. Of interest, abnormal short-term plasticity has been associated with deficits in moment-to-moment information processing and working memory, two hallmarks of schizophrenia (Crabtree and Gogos, 2014). Reduced efficacy in glutamate release may also be relevant for forms of long-term potentiation (LTP) that have clear presynaptic components, such as LTP at the perforant pathgranule cell synapse in the dentate gyrus (Errington et al., 1987). Notably, this form of LTP is impaired in DISC1∆2–3/∆2–3 mice—a stronger tetanic stimulus is required for the expression of LTP (Kuroda et al., 2011). Our findings suggest that this LTP deficit could be due to a failure of the presynaptic terminal to undergo activity-dependent increase in release probability.

Several genes encoding VGCC subunits, including CACNA1C, CACNB2 and CACNA1I have repeatedly been associated with schizophrenia and other psychiatric disorders (Ferreira et al., 2008; Cross-Disorder Group of the Psychiatric Genomics, 2013; Hamshere et al., 2013; Ripke et al., 2013; Schizophrenia Working Group of the Psychiatric Genomics, 2014). Although CACNA1A (Cav2.1) and CACNA1B (Cav2.2) are not typically associated with risk loci, RIM1 (also called RIMS1)—a presynaptic scaffold that regulates density of P/Q- and N-type Ca2<sup>+</sup> channels and SV docking at the active zone (Han et al., 2011)—was recently identified as a candidate gene for schizophrenia in the largest GWAS study conducted to date (Schizophrenia Working Group of the Psychiatric Genomics, 2014). Together these findings point to neurotransmitter release as a central process targeted in schizophrenia.

In conclusion, our results shed light on a novel mechanism by which a major susceptibility gene for mental illness enhances the efficacy of glutamate release, and provide further support for a central role of glutamate neurotransmission in schizophrenia and other major mood disorders.

### AUTHOR CONTRIBUTIONS

WT and JVT performed and analyzed all imaging and biochemical experiments. QL performed and analyzed whole-cell patch-clamp recordings. MB carried out the statistical analysis. KBL made the original observation implicating DISC1 in SV cycling. KKu and KKa made the DISC1 ∆2–3 mouse and the DISC1 C-terminus antibody. WT and MF wrote Matlab scripts. TWS and MF supervised the project. MF wrote the article.

### ACKNOWLEDGMENTS

We thank Bo Lu for critical reading of the manuscript. We are in debt to Durgadevi Alagappan and Tan Li-Ting for excellent technical assistance. This work was supported by a grant to MF from the Ministry of Education Academic Research Fund (MOE2013-T2-1-053).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnsyn. 2016.00015/abstract

FIGURE S1 | Variability of SV exocytic responses. Scatter plot showing SV exocytic rates in individual boutons across six (color-coded) independent preparations. Each preparation comprises three fields colored light to dark. These rates were measured in hippocampal neurons (DIV14–16) expressing scr shRNA and correspond to the data shown in Figure 2.

### REFERENCES


FIGURE S2 | Presynaptic localization and abundance of Cav2 channels are unaffected by DISC1 silencing. (A,D) Immunolocalization of endogenous Cav2.2 (A) or Cav2.1 (D) and bassoon in rat hippocampal neurons (DIV 14–15) expressing scr and DISC1-E shRNAs. Arrowheads indicate boutons positive for Cav2.2 and bassoon (A) or Cav2.1 and bassoon (D). Scale bar 10 µm. (B) Percentage of Cav2.2-positive boutons in scr (5 cells) and DISC1-E (5 cells) shRNA-expressing neurons. (C) Cav2.2 fluorescence intensity measured at individual presynaptic boutons (based on bassoon staining) in (B). (E) Percentage of Cav2.1-positive boutons in scr (8 cells) and DISC1-E (8 cells) shRNA-expressing neurons. (F) Cav2.1 fluorescence intensity measured at individual presynaptic boutons (based on bassoon staining) in (E). (G) Quantification of the density of presynaptic terminals (based on bassoon staining) in neurons expressing scr (24 cells), DISC1–E (21 cells) and -A (17 cells) shRNAs from four independent experiments.

TABLE T1 | Statistical analysis of synaptic responses. Column 1: datasets. Column 2: p values derived from the field averaging approach, with each field weighted according to its number of boutons (see "Materials and Methods" Section). Column 3: p values derived from the linear mixed model, where intra-field correlations and variations in neuron preparations are taken into account (see "Materials and Methods" Section).


presynaptic active zone. Neuron 69, 304–316. doi: 10.1016/j.neuron. 2010.12.014


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Tang, Thevathasan, Lin, Lim, Kuroda, Kaibuchi, Bilger, Soong and Fivaz. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Visualizing Presynaptic Calcium Dynamics and Vesicle Fusion with a Single Genetically Encoded Reporter at Individual Synapses

Rachel E. Jackson and Juan Burrone\*

Centre for Developmental Neurobiology, King's College London, London, UK

Synaptic transmission depends on the influx of calcium into the presynaptic compartment, which drives neurotransmitter release. Genetically encoded reporters are widely used tools to understand these processes, particularly pHluorin-based reporters that report vesicle exocytosis and endocytosis through pH dependent changes in fluorescence, and genetically encoded calcium indicators (GECIs) that exhibit changes in fluorescence upon binding to calcium. The recent expansion of the color palette of available indicators has made it possible to image multiple probes simultaneously within a cell. We have constructed a single molecule reporter capable of concurrent imaging of both presynaptic calcium influx and exocytosis, by fusion of sypHy, the vesicle associated protein synaptophysin containing a GFP-based pHluorin sensor, with the red-shifted GECI R-GECO1. Due to the fixed stoichiometry of the two probes, the ratio of the two responses can also be measured, providing an all optical correlate of the calcium dependence of release. Here, we have characterized stimulus-evoked sypHy-RGECO responses of hippocampal synapses in vitro, exploring the effects of different stimulus strengths and frequencies as well as variations in external calcium concentrations. By combining live sypHy-RGECO imaging with post hoc fixation and immunofluorescence, we have also investigated correlations between structural and functional properties of synapses.

#### Edited by:

Marc Fivaz, Duke-NUS Graduate Medical School, Singapore

#### Reviewed by:

Kirill Volynski, University College London, UK Michael Alan Cousin, University of Edinburgh, UK

#### \*Correspondence:

Juan Burrone juan.burrone@kcl.ac.uk

Received: 14 May 2016 Accepted: 13 July 2016 Published: 26 July 2016

#### Citation:

Jackson RE and Burrone J (2016) Visualizing Presynaptic Calcium Dynamics and Vesicle Fusion with a Single Genetically Encoded Reporter at Individual Synapses. Front. Synaptic Neurosci. 8:21. doi: 10.3389/fnsyn.2016.00021 Keywords: presynaptic, calcium, neurotransmitter release, pHluorin, RGECO, vesicle

### INTRODUCTION

Synaptic transmission depends on the influx of calcium into the presynaptic compartment, driving the release of vesicles containing neurotransmitter, which will bind to receptors on the postsynaptic cell. Electrophysiological studies have been fundamental in understanding the relationship between presynaptic calcium and neurotransmitter release (Dodge and Rahamimoff, 1967) but are limited in spatial resolution and cannot provide information about individual presynaptic terminals, unless employed at large synapses that can be accessed directly (Heidelberger and Matthews, 1992; Schneggenburger and Neher, 2000; Vyleta and Jonas, 2014). For studying individual synapses, optical strategies using either chemical indicators or genetically encoded reporters have therefore become the method of choice. Genetically encoded reporters offer the advantage that neural activity can be monitored in defined populations of cells, both in vitro and in vivo, and the probe can be localized to specific subcellular compartments (reviewed in Broussard et al., 2014).

The largest family of single color genetically encoded calcium indicators (GECIs) is the GCaMP family, based on a circularly permuted GFP linked to calmodulin (CaM) and it's binding peptide M13 (Nakai et al., 2001). The original sensor has undergone multiple rounds of both random and structure-guided mutagenesis to improve stability, brightness, signal to noise ratio and kinetics (Tian et al., 2009; Akerboom et al., 2012; Chen et al., 2013) and several variants have been effectively targeted to the neuronal presynaptic terminal by fusion with synaptophysin (Dreosti et al., 2009; Li et al., 2011; Nikolaou et al., 2012). To enable multicolor imaging in single cells, red-shifted GECIs have been developed, including RCaMP (Akerboom et al., 2013) and R-GECO1 (Zhao Y. et al., 2011), for which a presynaptically targeted version (SyRGECO) has also been characterized (Walker et al., 2013). Much like the GCaMPs, red GECIs are constantly evolving, with new variants recently produced in both RCaMP and R-GECO families (Inoue et al., 2015; Dana et al., 2016).

The most widely used reporters of synaptic vesicle exocytosis and endocytosis consist of a synaptic vesicle protein fused to pHluorin, a pH-sensitive form of GFP (Miesenbock et al., 1998). Crucially in these constructs, pHluorin is localized to the acidic lumen of synaptic vesicles, so that its fluorescence is quenched at rest, but increases by up to 20-fold upon fusion with the plasma membrane and exposure to the more basic pH of the extracellular medium (Sankaranarayanan et al., 2000). Several variants exist including fusions to the intraluminal domains of VAMP2, known as synaptopHluorin, (Miesenbock et al., 1998), VGLUT1 (Voglmaier et al., 2006), and synaptophysin, known as sypHy (Granseth et al., 2006). Recently, efforts have been made to improve the normalization of presynaptic pHluorin responses, which are complicated by the presence of a fraction of the protein at the cell surface, particularly for synaptopHluorin and sypHy. At rest, surface stranded protein dominates the baseline fluorescence due to its increased brightness compared to quenched protein localized to synaptic vesicles and is highly variable between boutons, preventing the use of baseline fluorescence as a normalizing factor for exocytic responses. By fusing tdimer2 to the C-terminus of sypHy Rose et al. (2013) generated ratio-sypHy, in which the red fluorescence provides an invariant signal proportional to the expression of the probe rather than the surface fraction. The ratio of green to red fluorescence is fixed and allows normalization of sypHy responses to the red baseline, simplifying comparisons across cells with different expression levels and at different depths within a tissue. Finally, red-shifted reporters of vesicle fusion have also been generated, including VGLUT-2XmOr2 (Li et al., 2011), synaptobrevin 2 mOrange (Ramirez et al., 2012), and sypHTomato (Li and Tsien, 2012), again offering the possibility of combined imaging with other spectrally distinct reporters.

Despite the development of these tools, relatively few studies have been carried out in which both presynaptic calcium influx and vesicle exocytosis have been imaged in the same small, enpassant presynaptic terminals. Ermolyuk et al. (2012) used FM dyes to monitor vesicle exocytosis in conjunction with a calcium indicator introduced via a patch pipette, to examine the major parameters of presynaptic function including the size of the RRP, the probability of single vesicle fusion, calcium influx and the overall release probability of the synapse. Whilst this elegant study was highly informative, the techniques used are limited to relatively small numbers of cells and cannot be transferred to an in vivo situation. With genetically encoded reporters, visualization of both calcium influx and neurotransmitter release has been achieved by co-expression of a red-shifted reporter of vesicle exocytosis, either VGLUT-mOr2 or sypHTomato, with a green calcium indicator, such as members of the GCaMP family (Li et al., 2011; Li and Tsien, 2012). However, in this situation the two probes were expressed separately, resulting in possible differences in the level of expression of each reporter and in their subcellular location. Crucially, it also precludes the use of a ratiometric approach to measuring sypHy responses (Rose et al., 2013), complicating the direct comparison of responses across cells.

We have generated a new genetically encoded probe that brings together the measure of calcium and of exocytosis, in addition to a ratiometric measure of sypHy signals. To do this we combined sypHy, a well characterized pHluorin sensor of synaptic vesicle exocytosis (Granseth et al., 2006), with the redshifted calcium indicator R-GECO1 (Zhao Y. et al., 2011) in a single molecule (sypHy-RGECO) to enable concurrent imaging of calcium influx and neurotransmitter release. The fixed 1:1 ratio of the two reporter molecules allows not only the normalization of sypHy signals to baseline or maximum RGECO fluorescence, but also the use of the ratio of the two responses as a measure of the calcium dependence of release at individual synapses. We systematically varied the number and frequency of action potentials, as well as the external calcium concentrations, and measured sypHy and RGECO signals in individual synapses. We found that while the amplitude of both responses changed, the distribution of the ratios remained constant. Interestingly, the calcium dependence of release was found to vary considerably between synapses, even within the same cell, suggesting the sensitivity of neurotransmitter release to calcium is synapsespecific. We also combined live sypHy-RGECO imaging with immunofluorescence for structural markers and observed that the amplitudes of calcium influx and exocytosis at individual synapses were positively correlated with the levels of the active zone protein RIM but were less well correlated with levels of the synaptic vesicle calcium sensor synaptotagmin-1.

### MATERIALS AND METHODS

### Generation of syn::sypHy-RGECO Plasmid

SypHy was amplified by PCR using primers containing HindIII and NotI restriction sites from CMV::sypHy A4 (a gift from Leon Lagnado, Addgene plasmid # 24478). The resulting PCR fragment was subcloned in to a modified version of the pEGFP-N2 plasmid (Clontech Laboratories, USA), in which the CMV promoter had been replaced with the human synapsin I promoter. To amplify RGECO from SyRGECO (Walker et al., 2013), a forward primer containing an XmaI site followed by a 5 amino acid linker (GSGGT) was used in conjunction with a reverse primer containing a NotI site. When subcloned into these

sites in the syn::sypHy-EGFP-N2 plasmid the EGFP was replaced to generate syn::sypHy-RGECO.

### Hippocampal Neuronal Culture and Transfection

Dissociated hippocampal neurons were prepared from Sprague– Dawley rats at embryonic day 18. Dissected hippocampi were treated with 5 mg/ml trypsin (Worthington, UK) for 5 min at 37 and triturated with fire-polished Pasteur pipettes. Neurons were plated on 18mm glass coverslips (Hecht Assistent, Germany) or 35 mm Grid500 µ-dishes (Ibidi, Germany) coated with poly-Dlysine (50 µg/ml) and laminin (20 µg/ml, both Sigma–Aldrich, UK). Cultures media were maintained in neurobasal media supplemented with 2% B27, 2% fetal bovine serum, 1% glutamax (all ThemoFisher Scientific, UK) and 1 % penicillin/streptomycin (Sigma), at 37◦C in a humidified incubator with 5% CO2. After 3 days in vitro (DIV) the media was exchanged for serum-free media. At 7DIV neurons were transfected with sypHy-RGECO using Effectene (QIAGEN, UK). After transfection neurons were maintained in serum-free media without antibiotics. Only schedule 1 procedures performed by a competent individual were used in these studies, which are exempt under the Animals (Scientific Procedures) Act 1986.

### Imaging of sypHy-RGECO

Neurons were imaged at 17–21DIV in HEPES buffered saline (HBS; 139 mM NaCl, 2.5 mM KCl, 10 mM HEPES, 10 mM D-Glucose, 2 mM CaCl2, 1.3 mM MgCl2; pH 7.3, 290 mOsmol) with 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (NBQX), 0.025 mM amino-5-phosphonovaleric acid (APV) and 6-imino-3-(4-methoxyphenyl)-1(6H)-pyridazinebutanoic acid hydrobromide (Gabazine) (all Tocris, UK). Coverslips were placed in a field stimulation chamber containing platinum electrodes (RC49-MFS, Warner Instruments, USA). For Ibidi dishes a stimulation insert (RC-37WS, Warner Instruments) was placed inside the dish. Neurons were imaged on an inverted Olympus IX71 microscope, equipped with a 60x/1.42 NA oil objective. Imaging was performed with a dual band-pass filter set optimized for EGFP and mCherry (Chroma, cat number 59022) and with LED excitation light sources of 470 and 585 nm (CoolLED, UK) for pHluorin and RGECO fluorophores, respectively. Time-lapse images were acquired at approximately 9.4 Hz using an Evolve 512 EMCCD camera (Photometrics, USA) controlled by Slidebook software (Intelligent Imaging Innovations, USA). Each frame consisted of a 20 ms exposure of the 470 and 585 nm LEDs sequentially, at 50 and 70% power, respectively, with 2 × 2 binning. Stimulation consisted of 1 ms 80 V pulses, which approximate single APs (Zhao C. et al., 2011), delivered by an SD9 Stimulator (Grass Technologies, USA), controlled by Slidebook. At the end of the imaging session 200 µM ionomycin was applied (Cayman Chemicals, USA) in either standard HBS, or HBS in which 50 mM of NaCl was replaced with 50 mM NH4Cl, to maximize sypHy-RGECO fluorescence. Neurons in Ibidi dishes were fixed in 4% PFA + 1% sucrose after imaging and were not treated with ionomycin. For calculations of surface fluorescence, synapses

### Analysis of Live Images

Images were analyzed using custom written Matlab codes (Mathworks, USA). Regions of interest (ROIs, 6 × 6 pixels) were selected for each puncta of sypHy fluorescence. Mean background-subtracted fluorescence intensity values were calculated for each ROI in both sypHy (G) and RGECO (R) channels. Traces were smoothed by averaging over a sliding window of 4 frames. Baseline fluorescence (G<sup>0</sup> and R0) was measured as the average of ten frames prior to the stimulus. 1G and 1R values were calculated by the change in signal intensity from the baseline, with the peak responses defined as the maximum 1G and 1R within 40 frames of the stimulus. Puncta in which the 1R response to the first 10 AP 20 Hz stimulus was greater than three times the standard deviation of the baseline were considered responding synapses and were analyzed in further images or stimulation conditions. 1G and 1R values for each synapse were averaged across trials and any synapse in which the mean peak response did not reach 3 SDs of the mean baseline in either channel were also discarded. Before pooling, mean 1G and 1R responses were normalized to either R<sup>0</sup> or Rmax, defined as the average of 10 images in 200 µM ionomycin.

### Immunofluorescence

After fixation cells were permeabilized with 0.2% Triton X-100 in PBS for 5 min, washed three times in PBS and placed in blocking solution (3% BSA in PBS + 0.05% NaN3) for 60 min at room temperature (RT), or for longer at 4◦C. Cells were incubated in primary antibodies diluted in blocking solution for 60 min at RT then washed five times in PBS. Secondary antibodies in blocking solution were applied for 60 min at RT and cells were washed five times in PBS. Cells were mounted using Ibidi mounting medium.

Cells were stained with chicken anti-GFP (1:1000, Abcam, UK) to amplify the sypHy signal, and either rabbit anti-RIM1/2 (1:500, Synaptic Systems, Germany) or mouse anti Synaptogmin-1 (1:100, Synaptic Systems). Appropriate secondary antibodies tagged with Alexa 488 or 647 (ThermoFisher Scientific) were used at 1:1000.

### Fixed Cell Imaging and Analysis

The regions previously imaged live were located using the grid coordinates and imaged on an inverted Nikon Eclipse Ti spectral confocal microscope equipped with a 60x/1.40 NA oil objective using NIS Elements software. Excitation wavelengths were 405, 488, and 636 nm with bandpass emission filters set to 425– 475 nm, 500–550 nm, and 662–737 nm. Microscope settings were kept the same for all images of the same antibodies. Z-stacks were generated from 0.15 µm optical sections, and maximum projections were produced in Image-J. To align the live and fixed images landmarks were selected in the GFP channel of both images, followed by affine transformation of the fixed image. ROIs from the live image were scaled and overlaid on the

registered GFP image for manual confirmation of the position. The GFP image was then thresholded using the 'Moments' algorithm in ImageJ to produce a mask, which was applied to the other channels of the fixed image to avoid the inclusion of neighboring puncta in the ROI. Intensity values for each ROI were then calculated from the masked images.

### Statistical Analysis

Mean responses are displayed ±SEM unless otherwise stated, and n = the total number of synapses. As data are not normally distributed, non-parametric statistical tests have been used.

### RESULTS

To generate a single molecule reporter of both presynaptic calcium influx and synaptic vesicle exocytosis, the R-GECO1 fragment was amplified from SyRGECO (Walker et al., 2013) and subcloned downstream of sypHy (Granseth et al., 2006) with a short linker sequence. The whole sequence was placed under control of the human synapsin I promoter to drive expression in neurons. The resultant syn::sypHy-R-GECO1 plasmid will hereafter be referred to as sypHy-RGECO (**Figure 1A**). With this probe, presynaptic calcium influx is reported by an increase in red fluorescence and exocytosis of synaptic vesicles by an increase in green fluorescence (**Figure 1A**). In dissociated hippocampal neurons transfected with sypHy-RGECO, punctate fluorescence in presynaptic boutons was observed in both green and red channels (**Figure 1B**). A field stimulation of 10 action potentials (APs) at 20 Hz was applied and individual boutons responded with increases in fluorescence in both channels (**Figure 1C**), the amplitudes of which (1G and 1R) were highly correlated (**Figure 1D**).

In order to pool responses from multiple cells or synapses, baseline fluorescence is often used to normalize reporter responses. However, for pHluorins this value can be dominated by protein localized to the plasma membrane, which is approximately 20-fold brighter than the quenched vesicular protein (Sankaranarayanan et al., 2000), and is highly variable between synapses. One method that has been used to overcome this problem involved fusion of the red fluorescent protein tdimer2 to the C-terminus of sypHy, generating a reporter termed ratio-sypHy (Rose et al., 2013). The fixed stoichiometry of the two fluorophores in this probe allowed normalization of the sypHy signals to the tdimer2 fluorescence. Using the same reasoning, RGECO can be used not only as a calcium indicator, but to normalize both 1G and 1R responses. As baseline RGECO fluorescence (R0) is relatively dim, we applied ionomycin in 2mM extracellular calcium to maximize the RGECO signal within the bouton (Rmax). Both values (R<sup>0</sup> and Rmax) were highly correlated (Supplementary Figure S1A) and were subsequently used as a normalizing factor for sypHy and RGECO responses. After ionomycin treatment, sypHy fluorescence also gradually increased to a plateau presumably due to vesicle release caused by the large influx of calcium. The maximum fluorescence in green and red channels was also highly correlated, reflecting the fixed ratio of the two fluorophores at all synapses (Supplementary Figure S1B). A similar correlation was observed when ionomycin treatment was combined with application of HBS containing 50mM NH4Cl to alkalinize all compartments and maximize sypHy fluorescence (Supplementary Figure S1C). This contrasts with the less tight correlations observed between the baseline fluorescence values of the two channels, G<sup>0</sup> and R<sup>0</sup> (Supplementary Figure S1D), and between G<sup>0</sup> and Gmax (Supplementary Figure S1E), which are likely affected by heterogeneity in the baseline fluorescence of sypHy. To directly measure the surface fraction of sypHy-RGECO, we compared the difference in sypHy fluorescence in external solutions at pH5.5 and pH7.4 to the sypHy fluorescence maximized by application of 50mM NH4Cl. The surface fraction varied between synapses (Supplementary Figure S1F, mean = 30.8%, median = 24.8%) and was slightly higher than that reported for ratio-sypHy (median f surf = 20%, Rose et al., 2013). This may be due to the fact that as basal RGECO fluorescence is low, it was necessary to identify transfected cells using basal sypHy fluorescence, potentially biasing toward brighter cells with higher surface fractions.

SypHy-RGECO responses to a 10 AP stimulus from multiple cells were normalized and pooled, and a positive non-linear correlation between 1G/Rmax and 1R/Rmax was observed (**Figure 1E**), as expected from the non-linear relationship between calcium and exocytosis (Dodge and Rahamimoff, 1967) and similar to that seen in other imaging studies (Zhao C. et al., 2011; Ariel et al., 2012; Ermolyuk et al., 2012). The spread of the responses suggested that individual synapses differed in the amount of neurotransmitter released for a given influx of calcium, which can be evaluated by the ratio 1G/1R. We observed a range of ratio values, the majority of which fell within a normal distribution, although a small number of synapses displayed very high release in comparison to calcium influx, an order of magnitude higher than the lowest ratios (**Figure 1F**). Thus, there is heterogeneity in the calcium dependence of release between synapses.

When using ratiometric imaging, differential bleaching of the fluorophores could pose a potential problem. In boutons imaged without stimulation, sypHy fluorescence decayed by only 2% during the imaging period. This decay could be fit with a double exponential, with time constants of 1.64 s and 3172 s (Supplementary Figure S2A). Frames within the first time constant were therefore disregarded and frames from 1 s before the stimulus were used to calculate baseline fluorescence. The slow phase of bleaching, which accounted for 98.5% of the decay, had a negligible effect on either baseline fluorescence or maximum 1G amplitude and was not corrected for, although this would need to be considered if imaging over a long period or using this probe to measure vesicular endocytosis rates. RGECO did not exhibit bleaching, instead showing an initial rapid increase in fluorescence that plateaued, the reasons for which will be discussed in more detail below (Supplementary Figure S2B). The rapid initial changes in fluorescence suggested that sequential images may be affected by differential bleaching. To test this, cells were stimulated with 10 AP stimuli five times, with a 1 min interval between images. The average 1G and 1R responses both decreased over the trials, with a slightly greater decrease in the calcium response (Supplementary Figure S2C).

merge of the two channels for the boxed region. (C) Changes in sypHy (i, 1G) and RGECO (ii, 1R) fluorescence in response to a field stimulus of 10 APs at 20 Hz. (n = 65 synapses from 1 coverslip, traces representing individual synapse responses shown in green/red, mean response of all synapses shown in black, stimulation period shown by black bar). (D) 1G and 1R responses are correlated (Spearman's rank correlation r<sup>s</sup> = 0.766, p < 0.001). (E) Cells were stimulated five times with a 10 AP 20 Hz stimulus. A mean response across trials was calculated and normalized to the maximum RGECO fluorescence in 200 µM ionomycin for each synapse. 1G/Rmax and 1R/Rmax responses from multiple cells show a non-linear correlation (r<sup>s</sup> = 0.727, p < 0.0001, n = 526 synapses, from 9 coverslips and 5 independent cultures). The black line shows a fit to a power function of the form axb+c, where b = 2.18. (F) Distribution of values for the ratio 1G/1R, a measure of the calcium dependence of release (for the same synapses as in E).

The reasons for these decreases are not clear, and may be different for the two responses. RGECO did not exhibit bleaching across trials, whilst the baseline fluorescence of sypHy decreased by approximately 20% (Supplementary Figure S2C). It has previously been suggested that GECIs targeted to the presynaptic compartment can exhibit rundown in responses (Walker et al., 2013), whereas the responses of synaptic pHluorins are similar across repeated trials (Burrone et al., 2006). However, the ratio of the two responses remained relatively stable across trials, increasing by only 9% between the first and fifth repeat (Supplementary Figure S2C). To avoid any inherent bias due to the effect of repeated trials we interspersed different stimulation conditions, such as number or frequency of AP stimuli.

It has previously been reported that R-GECO1 displays photoactivation when illuminated with 488 nm light (Akerboom et al., 2013), which may explain the increase in red fluorescence

observed in unstimulated boutons (Supplementary Figure S2B). To examine if this phenomenon affected sypHy-RGECO calcium responses, single channel images, illuminated with 585 nm light only, were compared to our usual imaging conditions, alternating 470 and 585 nm light. 1R responses were larger in images taken with 585 nm light only (Supplementary Figure S2C, upper trace), although when scaled for comparison, the kinetics of the response were the same in both imaging conditions (Supplementary Figure S2C, lower trace). The response amplitudes and baseline fluorescence in single channel and two channel images were highly correlated and, importantly, scaled linearly (Supplementary Figures S2D,E). These results suggested that in our imaging conditions, fast alterations between blue and green excitation light reduces the size of the RGECO responses in a multiplicative manner, without the introduction of non-linear artifacts.

To characterize the sypHy-RGECO response to different stimulus strengths, stimuli ranging from 1 AP to 20 AP at 20 Hz were delivered by field stimulation. Both responses increased linearly over the range of stimulus strengths tested (**Figures 2A,B**), as has been previously reported for SyRGECO (Walker et al., 2013). Peak 1G and 1R responses to different numbers of action potentials also showed a strong linear correlation (**Figure 2C**). The mean ratio between vesicle exocytosis and calcium influx did not differ between 5, 10, and 20 AP stimuli (**Figure 2D**). The mean ratio was significantly increased at lower stimulus strengths, although many synapses that responded at higher stimulus strengths fell below the detection threshold in the 1 AP and 2 AP stimuli, potentially biasing toward synapses with higher release probabilities in these groups. At higher numbers of APs, the longer stimulation times required would begin to overlap with the endocytosis and reacidification of synaptic vesicles, making the amplitude of sypHy responses harder to interpret. However, to check for potential saturation of the probe we also delivered 40 and 100 AP stimuli at 20 Hz. The calcium response indicated by RGECO was saturated at 40 APs and displayed no further increase in amplitude to a 100 AP stimulus, whilst sypHy responses continued to increase with longer stimulation trains (Supplementary Figure S3).

We next tested the effect of changing the external calcium concentration, recording sypHy-RGECO responses in both 0.5 mM and 2 mM calcium for the same set of synapses. As expected, both the calcium influx and level of vesicle release were significantly reduced in 0.5 mM calcium (**Figure 3A**), although the positive correlation between the two responses was maintained (**Figure 3B**). For synapses which responded above threshold in both conditions, the distribution of 1G/1R ratios was not significantly different between the two concentrations (**Figure 3C**), although the ratios measured for 0.5 mM calcium showed a skewed distribution toward higher values. This discrepancy is likely due to the saturation of the vesicular sensor of exocytosis for higher calcium concentrations, as reported previously (Ermolyuk et al., 2012).

Calcium influx at the presynaptic terminal is predominantly mediated by CaV2.1 and 2.2 channels, with some contribution from CaV2.3 (Takahashi and Momiyama, 1993; Wheeler et al., 1994; Gasparini et al., 2001). The contribution of different channel subtypes to the presynaptic calcium signal varies not only between synapses, but also with stimulation frequency (Ricoy and Frerking, 2014). We stimulated neurons expressing sypHy-RGECO with 10 APs at 5, 20, and 83 Hz, to examine whether stimulation frequency affected the relationship between 1G and 1R responses. Pooled 1G and 1R responses across all frequencies showed the expected non-linear relationship (**Figure 3D**). Both calcium influx and vesicle exocytosis were significantly reduced at 83 Hz stimulation, which could be due to the failure of the cell to fire action potentials at high frequency, or to the inactivation of calcium channels (**Figures 3D,E**). However, the distributions of 1G/1R ratios were indistinguishable between stimulation frequencies (**Figure 3F**), suggesting frequency did not alter the relationship between calcium influx and vesicle release.

Synaptic boutons are heterogeneous in both structural and functional parameters. Studies of the relationship between the two show that bouton volume is a poor predictor of functional properties such as presynaptic calcium influx and the synaptic probability of release (Ermolyuk et al., 2012; Holderith et al., 2012), whereas active zone size strongly correlates with function (Holderith et al., 2012). Additionally, levels of the active zone proteins RIM1/2 are highly correlated with active zone size (Holderith et al., 2012). To examine the relationship between sypHy-RGECO responses and active zone size, neurons were first imaged live, whilst stimulating with 10 APs at 20 Hz, and subsequently fixed and stained for RIM1/2. Live and fixed images were aligned to allow the comparison of functional and structural data from individual identified synapses (**Figure 4A** and Supplementary Figure S4). We observed significant positive correlations between RIM levels and both sypHy and RGECO responses (**Figure 4Bi,ii**). Whilst significant, these correlations are weaker than those found in ultrastructural studies, most likely due to limitations in the alignment process and the lower resolution of light microscopy. Interestingly, RIM levels also correlated significantly, albeit weakly, with the ratio 1G/1R (**Figure 4Biii**), possibly reflecting the role of RIM, along with RIM binding protein (RBP), in localizing calcium channels to the active zone (Kaeser et al., 2011), vesicle docking (Han et al., 2011) and coupling of calcium channels to release (Kaeser et al., 2012).

Similar correlations were also carried out with staining for the putative calcium sensor for synchronous neurotransmitter release, synaptotagmin-1 (Syt1) (Geppert et al., 1994; Fernandez-Chacon et al., 2001). We found that the levels of Syt1 were positively correlated with the magnitude of the RGECO response, and showed a trend toward a positive correlation with neurotransmitter release that did not reach significance (**Figures 4Ci,ii**). However, the ratio of release to calcium influx at each synapse was not related to the level of Syt1 (**Figure 4Ciii**). As the neurons were stimulated with a train of APs at high frequency (20 Hz), both synchronous and asynchronous forms of release are expected to operate (Hagler and Goda, 2001), and thus the levels of other calcium sensors may be more important in determining the total release that occurs in response to this type of stimulus. Furthermore, approximately

synapses from 4 coverslips and two independent cultures.

20% of total Syt1 is found on the surface of the bouton, not on vesicles (Fernandez-Alfonso et al., 2006), thus the level of Syt1 present at the bouton may not accurately reflect the size of the vesicle pool. Together, our data is in broad agreement with the general principle that synapses that have larger active zones also show higher levels of calcium influx and neurotransmitter release. Although the overall levels of Syt1 at the synapse also broadly correlated with these functional measures, they were a worse predictor of synapse function than RIM.

### DISCUSSION

We have generated a single molecule reporter, sypHy-RGECO, for concurrent imaging of presynaptic calcium influx and synaptic vesicle recycling. This hybrid molecule consisted of a red calcium indicator, R-GECO1 (Zhao Y. et al., 2011), fused to the C-terminus of sypHy, a reporter of vesicle exocytosis and endocytosis (Granseth et al., 2006). Combining the two reporters in one molecule offers several advantages. First, it provides two independent readouts of presynaptic function, without the need for co-expression of multiple probes. Second, as the probes are not separated spatially and a fixed ratio of the two fluorophores is present at all synapses, it provides a way of normalizing sypHy signals across synapses and cells, in a manner similar to that employed for ratio-sypHy (Rose et al., 2013). This is particularly important for sypHy signals, where resting levels of fluorescence vary quite dramatically from bouton to bouton and do not correlate directly with the amount probe at the synapse, but rather to its surface levels. Finally, the fixed stoichiometry also allows a direct comparison of the two fluorescence responses, thereby providing a direct measure of the calcium dependence of release within single presynaptic boutons.

FIGURE 3 | Effects of altering stimulation conditions. (A–C) Dissociated hippocampal neurons in external solution (HBS) containing 0.5 mM calcium were stimulated with a train of 10 APs at 20 Hz repeated five times 1 min apart. The solution was replaced with HBS containing 2 mM calcium and the stimulus set repeated. Responses across trials were averaged for each synapse. (A) Fluorescence changes in sypHy (upper traces) and RGECO (lower traces) in 0.5 and 2 mM. Traces show mean ± SEM for n = 131 synapses from 3 coverslips from 3 independent cultures. (B) Correlation between peak 1G/Rmax and 1R/Rmax responses for these synapses (Spearman's rank correlation r<sup>s</sup> = 0.873, p < 0.001, the black line indicates a fit to a power function of the form axb+c, where b = 1.30.) (C) Distribution of 1G/1R ratios for synapses which responded above threshold at both calcium concentrations, ratios shown are averaged from the last two 0.5 mM trials (gray) and the first two 2 mM trials (black) (n = 47 synapses). The distribution is not significantly different between concentrations (Kolmogorov–Smirnov test, p = 0.093). (D–F) Hippocampal neurons were stimulated with 10 APs at 5, 20, and 83 Hz. Each set of stimuli was repeated three times, varying the order in which the frequencies were delivered and the mean of the trials was calculated. (D) Correlation between peak 1G/Rmax and 1R/Rmax values at 5 Hz (blue), 20 Hz (red), and 83 Hz (green). Mean responses to each frequency are shown in the larger, brighter points (n = 134 synapses from 3 coverslips and three independent cultures, the black line indicates a fit to a power function of the form axb+c, where b = 1.91). (E) Distribution of 1G/Rmax (left) and 1R/Rmax (right) responses for these synapses show a decrease in the responses at 83 Hz compared to 20 Hz (Kolmogorov–Smirnov test, 1G/Rmax p = 0.0013, 1R/Rmax p < 0.001) but no change in the response at 5 Hz (Kolmogorov–Smirnov test 1G/Rmax p = 0.537, 1R/Rmax p = 0.240). (F) The distribution of 1G/1R ratios does not change between stimulus frequencies (Kolmogorov–Smirnov test 5 Hz p = 0.703, 83 Hz p = 0.282).

We observed a strong non-linear correlation between calcium influx and vesicle exocytosis across synapses that was maintained in response to different stimulation conditions, in agreement with many previous studies (Ariel et al., 2012; Ermolyuk et al., 2012). Using the ratio of the sypHy response to the RGECO response (1G/1R), we were also able to measure the calcium-dependence of release at individual presynaptic boutons, a feature that has been difficult to study in these small CNS synapses. Interestingly, we observed heterogeneity in this value, suggesting that calcium triggers vesicle fusion with different efficiencies at different synapses. It is not yet clear what underlies this variation, as the distribution of 1G/1R ratios did not change significantly under different stimulation conditions, such as stimulation frequency and extracellular calcium, despite changes in the amplitude of both calcium and exocytosis signals. A possible source of this variation may be the repertoire of presynaptic calcium channel subtypes, which can vary between synapses (Reid et al., 1997; Nimmervoll et al., 2013). In support of this, individual boutons displayed differential sensitivity to pharmacological blockade of CaV2.1 and 2.2 channels in both calcium influx measured by GCaMP3 and exocytosis measured by sypHTomato, when these reporters were co-expressed in the same cell (Li and Tsien, 2012). Similarly, in another study in which calcium and vesicle dynamics were imaged in separate cells, pharmacological inhibition of CaV2.1 or 2.2 led to variable blockade of neurotransmitter release between individual synapses, although this did not alter the relationship between calcium and exocytosis measured across the synapse population (Ariel et al., 2012). However, neither

independent cultures). (C) Levels of Syt1 are not significantly correlated with sypHy responses (i, p = 0.059), or the 1G/1R ratio (iii, p = 0.323), but are correlated

of these studies directly examined the relationship between channel subtypes and the calcium dependence of release in individual boutons, which could be addressed using sypHy-RGECO.

with the calcium response (ii, p = 0.005) (n = 62 synapses from three dishes from independent cultures).

An alternative possibility is that individual boutons differ in the levels or components of the molecular machinery that links calcium influx to vesicle release. Of the markers that we examined using post hoc immunofluorescence, the level of the

calcium sensor synaptotagmin-1 was not related to the 1G/1R ratio, although a weak correlation was observed with the level of RIM, an active zone protein with roles in the localization of calcium channels and vesicle docking (Han et al., 2011). Whilst absolute RIM levels are only weakly predictive of the ratio, it is possible that more nuanced features such as the size and shape of the active zone relative to the bouton volume, and positioning of elements such as calcium channels within it may affect the calcium dependence of release at individual synapses. Combining functional imaging with structural imaging by immunofluorescence, super-resolution, or electron microscopy will enable investigation of these possibilities. Such methods could also be used to examine if differences in the calcium dependence of release exist in different cell types, or to study the relationship between presynaptic function and postsynaptic properties at individual connections.

Finally, it would be interesting to examine if the calcium dependence of release is fixed for an individual synapse or is subject to regulation. Calcium influx and vesicle exocytosis have been measured individually after chronic changes in network activity, and the changes observed indicate that the relationship between the two is not altered at the population level following homeostatic plasticity (Zhao C. et al., 2011). However, this is yet to be studied simultaneously in single synapses. Furthermore, it is unknown if short-term plasticity mechanisms involve changes in the efficiency with which calcium triggers vesicle fusion. Tools such as sypHy-RGECO will enable optical interrogation of these questions.

Several spectrally distinct genetically encoded reporters of calcium and vesicle recycling have been developed, offering the opportunity to combine probes with different characteristics. We selected to combine sypHy and R-GECO1 for several reasons; synaptophysin has previously been shown to tolerate addition of both pHluorin in the intravesicular loop with a second fluorophore at the C-terminus (Rose et al., 2013), and both sypHy and SyRGECO have previously been characterized individually (Granseth et al., 2006; Walker et al., 2013). SypHy has a superior signal-to-noise ratio than synaptopHluorin and whilst VGLUT-pHluorin offers further improvement in this aspect (Voglmaier et al., 2006), there is a risk that VGLUT1 overexpression may cause an increase in quantal content (Wojcik et al., 2004; Wilson et al., 2005), and therefore alter the strength of synaptic connections. Additionally, although VGLUT based probes have been used in inhibitory neurons, it may impart GABAergic boutons with unwanted properties resulting from the overexpression of a transporter not usually found in these neurons. R-GECO1 is a sensitive calcium indicator with a wide dynamic range that exhibits a larger fluorescence change to field stimuli than the RCaMP1 sensors (Akerboom et al., 2013; Walker et al., 2013). RCaMP2 (Inoue et al., 2015) offers greater sensitivity than R-GECO1 but saturates at approximately 8 APS and was therefore less suitable for combination with sypHy, for which longer stimulus trains are typically used. The 1F/F responses of R-GECO1 are also linear over a greater range of calcium concentrations than RCaMP1 sensors (Akerboom et al., 2013). When combined in the single molecule sypHy-RGECO, both reporters displayed a linear increase in the amplitude of the response across the range of 1–20 AP stimuli delivered at 20 Hz, but at higher stimulus strengths the RGECO response became saturated. Whilst sypHy responses continued to increase to both 40 and 100 AP stimuli, the amplitude of these responses will be affected by the decrease in fluorescence caused by endocytosis and reacidifcation of vesicles. The dynamic ranges of the two reporters are therefore well matched and suited to reporting activity up to 20 action potentials.

One potential disadvantage of combining R-GECO1 with green sensors of vesicle exocytosis is its photoswitching behavior in response to 488 nm light (Akerboom et al., 2013). Indeed, we found that blue light exposure in our imaging conditions did cause an offset in the basal RGECO fluorescence, as well as a reduction in the calcium response. However, these decreases scaled linearly with the original R-GECO1 response, and both the kinetics of the calcium transient and the range of responses across synapses were preserved. SypHy-RGECO is therefore a useful addition to the expanding toolbox of genetically encoded reporters, and will enable a detailed study of the relationship between calcium influx and vesicle dynamics at individual synapses. Whilst we have shown proof of principle in hippocampal synapses in vitro, this reporter could also be used to study these processes in more intact systems, such as brain slices and in vivo.

### AUTHOR CONTRIBUTIONS

JB and RJ designed experiments, RJ performed experiments, analyzed data, and drafted the manuscript, JB and RJ edited and approved the manuscript.

### FUNDING

This work was funded by an ERC consolidator grant, a Wellcome Trust Investigator award and a Lister Institute Prize to JB.

### ACKNOWLEDGMENTS

We would like to thank Mideia Kotsogianni for the preparation of dissociated hippocampal cultures, Robert E. Campbell for the R-GECO1 plasmid (Addgene plasmid 32444) and Leon Lagnado for the sypHy plasmid (Addgene plasmid 24478).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnsyn. 2016.00021

### REFERENCES

fnsyn-08-00021 July 25, 2016 Time: 16:6 # 11



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Jackson and Burrone. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Flash-and-Freeze: Coordinating Optogenetic Stimulation with Rapid Freezing to Visualize Membrane Dynamics at Synapses with Millisecond Resolution

#### Shigeki Watanabe1, <sup>2</sup> \*

*<sup>1</sup> Department of Cell Biology, Johns Hopkins University, Baltimore, MD, USA, <sup>2</sup> The Solomon H. Snyder Department of Neuroscience, Johns Hopkins University, Baltimore, MD, USA*

Electron microscopy depicts subcellular structures at synapses exquisitely but only captures static images. To visualize membrane dynamics, we have developed a novel technique, called flash-and-freeze, which induces neuronal activity with a flash of light and captures the membrane dynamics by rapid freezing. For characterizing membrane movements during synaptic transmission, a light-sensitive cation channel, channelrhodopsin, is heterologously expressed in mouse hippocampal neurons or in *Caenorhabditis elegans* motor neurons. A brief pulse of blue light activates channelrhodopsin and induces an action potential, leading to synaptic transmission. Following the light stimulation, neurons are frozen at different time intervals ranging from 10 ms to 20 s. Electron micrographs are then acquired from each time point to visualize the morphological changes. Using this approach, we have characterized a novel form of endocytosis, ultrafast endocytosis, which rapidly removes excess membrane added to the surface during neurotransmission. The flash-and-freeze approach can be adapted to study other cellular phenomena that can be induced by light-sensitive genetic or pharmacological tools.

Keywords: flash-and-freeze, synaptic transmission, synaptic cell biology, synaptic vesicle exocytosis, synaptic vesicles, optogenetics, high-pressure freezing, electron microscopy

### INTRODUCTION

### Imaging Neurons

To process information at synapses, membrane trafficking takes place on a millisecond time scale. When a nerve impulse arrives at synaptic terminals, synaptic vesicles fuse with the plasma membrane within 1 ms (Heuser and Reese, 1981). Subsequently, vesicle membrane and proteins are recovered from the plasma membrane to reconstitute new vesicles (Dittman and Ryan, 2009; Saheki and De Camilli, 2012). One major challenge in studying membrane dynamics at synapses is visualizing these rapid events at the spatial resolution necessary to resolve individual vesicles and temporal resolution relevant to synaptic functions. Several methods have been developed over the years including electrophysiological, optical, and electron microscopy methods.

Capacitance measurements of synaptic terminals allow monitoring of membrane flux at the plasma membrane with sub-millisecond temporal resolution (Hamill et al., 1981; Neher and Marty, 1982). The capacitance across a membrane is proportional to the membrane area.

#### Edited by:

*George Augustine, Nanyang Technologicial University, Singapore*

#### Reviewed by:

*Tom Reese, National Institutes of Health, USA Ruud Toonen, Vrije Universiteit Amsterdam, Netherlands*

\*Correspondence:

*Shigeki Watanabe shigeki.watanabe@jhmi.edu*

Received: *25 May 2016* Accepted: *03 August 2016* Published: *19 August 2016*

#### Citation:

*Watanabe S (2016) Flash-and-Freeze: Coordinating Optogenetic Stimulation with Rapid Freezing to Visualize Membrane Dynamics at Synapses with Millisecond Resolution. Front. Synaptic Neurosci. 8:24. doi: 10.3389/fnsyn.2016.00024* Direct patch-clamping of synaptic boutons shows an increase in the capacitance during exocytosis and a decrease during endocytosis (von Gersdorff and Matthews, 1994; Sun et al., 2002; Delvendahl et al., 2016). This method has a single vesicle sensitivity. However, it is blind to the locations of the membrane insertion and removal. Furthermore, membrane trafficking within the cytosol cannot be visualized by this technique.

Optical imaging methods aim to label functional pools of vesicles in the terminals and visualize their fates. The spatial resolution of optical microscopy is limited to ∼200 nm by the diffraction of light. Because synaptic vesicles are small (∼40 nm in diameter) and hundreds of these vesicles are clustered in a confined space, optical microscopy cannot resolve individual vesicles in the terminal. Therefore, several tricks have been developed to visualize specific subsets of vesicles. For example, fluid phase markers like dextran, ferritin, and quantum dots can be applied exogenously to label endocytosed vesicles and visualize their dynamics in the terminal (Zhang et al., 2007b; Park et al., 2012). Similarly, lipophilic dyes like FM dyes and mCling can be internalized via endocytosis and mark the newly reconstituted vesicles (Betz et al., 1992; Revelo et al., 2014). Finally, pH-sensitive fluorescent molecules such as pHluorin (Miesenböck et al., 1998) can be used to monitor synaptic vesicle cycle. Fluorescence from pHlourin is quenched in the lumen of synaptic vesicles due to the low pH. When exposed to the extracellular space by exocytosis, the protein becomes fluorescent. The pHluorin molecules are quenched again once they are internalized and vesicles are fully re-acidified. These techniques allow visualization of particular vesicles (Sankaranarayanan and Ryan, 2000; Balaji and Ryan, 2007; Armbruster et al., 2013). However, like in the capacitance measurement, it is difficult to visualize intracellular trafficking events, limiting the interpretation of the results.

Electron microscopy, on the other hand, depicts all the membrane-bound structures within the terminal, but only captures a static image. To overcome this limitation, neurons are typically stimulated and fixed at defined time points after the stimulation (Ceccarelli et al., 1972; Heuser and Reese, 1973; Ferguson et al., 2007; Clayton et al., 2008; Matthews and Sterling, 2008; Hoopmann et al., 2010; Schikorski, 2014; Wu et al., 2014). However, the temporal resolution is limited to seconds to minutes for two reasons. First, to ensure the capture of endocytic events, neurons are often stimulated with prolonged intense stimulation that lasts seconds to minutes. Second, diffusion and reaction of aldehyde-based chemicals are slow, requiring an additional time. To visualize membrane dynamics on a millisecond time scale, Heuser and Reese developed the "freeze slammer," which freezes tissues near instantaneously following electrical stimulation (Heuser et al., 1979; Heuser and Reese, 1981). Using this device, they were able to observe vesicles fusing with the plasma membrane 5 ms after an electrical pulse (Heuser and Reese, 1981; Heuser et al., 1979). Although individual images are static, membrane dynamics can be visualized in electron micrographs by inducing a particular activity and freezing neurons at multiple defined time points.

The freeze slamming approach allows visualization of membrane dynamics at the temporal resolution relevant to synaptic functions, but it has two major drawbacks. First, the specimen must be physically attached to the electrical wire, and thus experiments can only be performed ex vivo. Second, only the surface of specimens (∼5 µm) can be frozen without ice crystal damage. These two factors limit its utility as a versatile tool for studying membrane trafficking at synapses. To overcome these limitations, we have developed a novel technique, flash-andfreeze, that combines optogenetics with high-pressure freezing (Watanabe et al., 2013a,b, 2014a). With this technique, noninvasive light stimulation is used to induce synaptic transmission. Specimens as thick as 200 µm can be properly frozen without ice crystal formation. Therefore, intact animals like Caenorhabditis elegans (C. elegans hereafter) or an entire population of neurons in a dish can be studied. Here, I will describe the methods in detail and discuss its potential applications in synaptic cell biology.

### High-Pressure Freezing

An electron microscope operates under high vacuum to avoid the scattering of electrons by gaseous molecules in the air. Thus, to observe a specimen in a transmission electron microscope, it must be fixed and dehydrated. In addition, electron beam must penetrate through the specimen, requiring extremely flat specimens with a thickness of ∼30–70 nm. For this reason, the specimen is embedded in plastic and sectioned ultrathin. The sample preparation for electron microscopy often leads to the generation of artifacts. Fixation using aldehyde-based chemicals cross-links proteins and aggregates them. Worse yet, this reaction can induce fusion of synaptic vesicles (Smith and Reese, 1980). Furthermore, dehydration leads to the shrinkage of the membrane-bound structures and the overall changes in the morphological architecture of the cells. Therefore, a better approach must be used to study membrane trafficking events at synapses.

One approach to avoid these artifacts is to immobilize cells physically by rapid freezing. The freezing process, however, leads to formation of ice crystals that can damage the cellular architecture by directly penetrating through the membrane. Alternatively, the solutes segregated from ice crystals can burst membrane due to the local changes in the osmotic pressure. To prevent water molecules from forming ice crystals, a freezing rate of at least 10,000 K/s must be achieved. At this rate, water molecules cannot rearrange to form ice crystals and are frozen in an unordered state. The cooling rate by liquid nitrogen can exceed 16,000 K/s. Unfortunately, the heat conductance of water is poor, reducing the rate to 1000 K/s within 10 µm from the point of the contact. Under high pressure (2100 bar), however, a freezing rate of 100 K/s is sufficient to freeze water in an unordered state due to the supercooling effect (Moor, 1987; Dubochet, 2007). Thus, high-pressure freezing allows freezing of tissues up to 200 µm thickness or intact animals like C. elegans, (∼70 µm thick) without ice crystal damage.

### Optogenetics

To visualize membrane dynamics during neurotransmission, neurons must be stimulated and frozen at defined time points after stimulation. In the freeze slammer, specimens are physically attached to the stimulating wire, limiting the specimen choice. Watanabe Flash-and-Freeze

Furthermore, the configuration of the high-pressure freezer makes it difficult to apply electrical field stimulation. To overcome these limitations, we have coupled optogenetics with high-pressure freezing. Channelrhodopsin is a light-sensitive cation channel isolated from Chlamydomonas reinhardtii (Nagel et al., 2003). A flash of blue light opens the channel, allowing cations to flow into the cell, thereby depolarizing it. When the channel is heterologously expressed in neurons, a short pulse of light triggers an action potential, leading to synaptic transmission (Boyden et al., 2005; Nagel et al., 2005). Therefore, non-invasive stimulation can be applied to a population of neurons in a dish or intact animals.

To couple optogenetic stimulation with high-pressure freezing ("flash-and-freeze"), we have developed a device that interfaces with the computer of high-pressure freezer as well as with an LED (Watanabe et al., 2013a,b, 2014a). This device allows application of light pulses at defined time points before the specimen is frozen (see Section Materials and Methods). Using this approach, we can visualize the membrane trafficking events at synapses with a millisecond temporal resolution.

### MATERIALS AND METHODS

All experiments are performed according to the guidelines for the animal use by the National Institute of Health. The animal protocol is approved by the Animal Care and Use Committee at Johns Hopkins University, School of Medicine. The graphical representations of the workflow is shown in **Figure 1**. The stepby-step protocol can be found in the Supplementary information.

### Cell Cultures

For flash-and-freeze experiments, specimens must be mounted on a substrate that is translucent like a glass coverslip but can withstand the extreme pressure application. Sapphire disks are ideal substrates, meeting these two criteria. In addition, thermal conductivity of sapphire is extremely high, particularly at low temperature, making it the perfect substrate. For culturing cells on sapphire disks, the following procedures are performed. First, we sputter carbon on one side of 6 mm sapphire disks so that we can distinguish the surface that the cells are cultured on. Then, we scratch out a letter "4" on the carbon-coated surface with a diamond scribe. Alternatively, a finder grid can be placed on a sapphire disk as a mask. The sapphire disks are baked at 120◦C overnight. After a brief dip in ethanol, two sapphire disks are placed in each well of a 12-well plate and air-dried in a biosafety cabinet. Poly-D-Lysine solution (acetic acid 3 parts, rat tail collagen 1 part, and poly-D-lysine 1 part) is applied directly on each sapphire disk. After 5 min, the excess solution is removed, and the plates are dried under the laminar flow. Prior to use, the 12-well plates containing sapphire disks are sterilized by exposure to UV light for 20 min.

For hippocampal cultures, we first culture astrocytes as a feeder layer and then plate the neurons on top of the astrocytes. Mouse brains are dissected out from newborn C57/BL6J mice immediately after decapitation. Cortices and hippocampi are isolated from each brain, in cold HBSS solution. For the astrocyte feeder layer, cortices are treated with 0.05% Trypsin-EDTA for 20 min at 37◦C. The dissociated astrocytes are then cultured in DMEM containing 10% FBS and 0.1% penicillin-streptomycin for 2 weeks in a T-75 flask. Then astrocytes are transferred to the 12-well plate containing sapphire disks (5 × 10<sup>4</sup> / well) and grown for a week. Astrocyte mitosis is arrested a few hours before neuron plating using fluorodeoxyuridine (80 mm). Media are exchanged with Neurobasal, a media containing 2% B27, 1% glutamax, and 0.1% penicillin-streptomycin, prior to neuron plating. Isolated hippocampi are treated with papain (20 U/ ml) for 1 h and plated (5 × 10<sup>4</sup> /well) on top of the feeder layer. Neurons are allowed to grow for 2–3 weeks. On DIV1-3, neurons are infected with lentivirus containing Channelrhodopsin expressed from the neuron specific synapsin promoter.

### Caenorhabditis elegans

Transgenic animals expressing ChIEF, a variant of channelrhodopsin-2, are grown on an agar plate (50 mm) seeded with 250 µl of E. coli OP50. Larval stage 4 (L4) transgenic animals are transferred to agar plates seeded with bacteria and 4 µl trans-retinal solution (100 mM) a day prior to the experiment. These animals are kept in the dark before the experiments.

### Freezing

We have used three different high-pressure freezers: Leica EM PACT2, HPM100, and ICE. C. elegans has been tested in the EM PACT2, whereas neuronal cultures are frozen with the HPM100 or ICE. For the EM PACT2, a membrane carrier with a 100 µmdeep well is used to mount transgenic worms (**Figure 2A**). First, the well of the membrane carrier is filled with M9 worm buffer containing 20% bovine serum albumin (BSA) as cryo-protectant. Approximately 10–15 young adult animals are transferred into the solution. The membrane carrier is then mounted into the modified specimen pod.

To mount cultured neuronal cells, the following procedures are carried out (**Figure 2B**). First, a sapphire disk with neurons is transferred into pre-warmed (37◦C) recording solution containing 140 mM NaCl, 2.4 mM KCl, 10 mM HEPES (pH 7.5), 10 mM glucose, 4 mM CaCl2, 1 mM MgCl2, 3 µM NBQX, and 30 µM bicuculline. The addition of NBQX and bicuculline blocks the recurrent network activity following channelrhodopsin stimulation. The sapphire disk is mounted on the countersink of a 6-mm middle plate. A spacer ring (100 µm thickness) is placed on the sapphire disk so that cells are not crushed when the assembly is capped with another sapphire disk. To cap the assembly, a sapphire disk is first dipped in the recording solution so that one side of the disk carries a drop of solution. This sapphire disk is then placed on top of the spacer ring gently so that air bubbles are not trapped in between the disks. Two more spacer rings (200 µm each) are then placed to fill up the remaining space. The excess solution should be removed by gently tapping with a small piece of filter paper. The middle plate assembly is then sandwiched between two transparent plastic half cylinders. Note that light is shone

from the bottom side of the specimen assembly in the HPM100 while it is from the top in the ICE. The video demonstrating sample loading can be found on the Leica website (http://www.leica-microsystems.com/science-lab/videotutorials-filling-and-assembling-of-different-carriers-for-highpressure-freezing/).

### Programming Light Stimulation

The modifications we have made for the EM PACT2 are described in detail in previous publications (Watanabe et al., 2013a,b, 2014a). For controlling the light stimulation and freezing, we constructed an Arudino Uno based device that sends out 5V TTL signals to the LED and the high-pressure freezer at the

desired time point. This device is driven by custom firmware to produce any pattern of light pulses (i.e., single stimulus, 10 Hz stimulation) and send out the "start" signal that triggers the freezing process (this device can be purchased from Marine Reef International). Once the signal is sent, the freezer applies hydraulic fluid to the specimen and pressurize the specimen. When the pressure reaches 2000 bar, the liquid nitrogen is immediately applied to the specimen. The initiation of this freezing process needs to be timed so that the desired time interval between the light stimulation and freezing is achieved. To monitor the precise timing when the pressure is applied to the specimen, we installed an accelerometer on the sample holder (also known as a "specimen bayonet"). When the pressure is applied to the specimen, the sample holder jolts, producing a distinctive peak in the accelerometer recording. We found that it requires about 170 ms for the EM PACT2 to apply the pressure to the specimen after sending the 'start' signal. After the pressure is applied to the specimen, the specimen is frozen to −20◦C in 8 ms, according to standard thermodynamic equations. To account for the variability in timing, we calculate the actual time interval posthoc based on the pressure peak recorded by the accelerometer. We have found the actual time interval to typically be within 33.6 ± 4.6 ms of the intended time.

To introduce light stimulation capability in the HPM100, a similar Arudino Uno based device was constructed. The operation of the HPM100 is fundamentally different from that of the EM PACT2 in that pressurized liquid nitrogen is applied to the specimen instead of a hydraulic fluid. After sending the "start" signal, it takes 370 ms to compress the liquid nitrogen. The pressurized liquid nitrogen reaches the specimen in precisely 72 ms. The specimen is frozen to −20◦C in 8 ms from the time the pressurized liquid nitrogen hits the specimen surface. Therefore, a total of 450 ms delay is expected. To monitor the precise timing, light stimulation device record the internal signal of the machine that opens the liquid nitrogen valve in each shot. This signal initiates almost invariably at 370 ms after the start signal. The actual timing of the freeze is adjusted post-hoc based on the valve opening signal.

For EM ICE, Leica microsystems has integrated the light stimulation control into the freezer. The machine is capable of freezing specimen at the desired timing after the light stimulation with 1–2 ms variability.

### Freeze-Substitution

Following high-pressure freezing, vitrified water from the specimen must be substituted with organic solvent to avoid crystallization of vitrified water as the specimen warms up to room temperature for further processing. The freeze-substitution is carried out in cryotubes containing fixatives (1% osmium tetroxide, 1% glutaraldehyde, 1% milliQ water in anhydrous acetone) using an automated freeze-substitution unit (Leica AFS2). The cryotubes containing fixatives are stored under liquid nitrogen to avoid the cross-reaction between osmium tetroxide and glutaraldehyde. The sapphire disks containing the cells typically remain associated with the middle plate. To release the sapphire disks, the middle plate is quickly transferred to a cup containing acetone, which is precooled to −90◦C in AFS. Once the middle plate reaches −90◦C in the acetone, the sapphire disk often dissociates from the middle plate spontaneously. If not, it can be released by a gentle tap. The sapphire disk is then transferred into the cryotube containing the fixatives at −90◦C. For C. elegans, the frozen specimens can be transferred under liquid nitrogen into the cryotubes containing fixatives. The tubes should be transferred quickly into the AFS. The following program is used for the freeze-substitution: −90◦C for 5–30 h, 5 ◦C/h to −20◦C, 12 h at −20◦C, 10◦C/h to 20◦C. The cryotubes are agitated at least twice a day during the substitution process.

### Plastic Embedding

Once the freeze-substitution is complete, specimens are embedded in Epon/Araldite resin. The resin is as composed of 4.4 g Araldite 502, 6.2 g Eponate 12, 12.2 g dodecenyl succinic anhydrite (DDSA), and 0.8 ml benzyldimethylamine (BDMA). First, the fixative solution is carefully removed from the cryotube, and specimens are washed with acetone a few times. Specimens are then treated with 0.1% uranyl acetate in acetone for 1 h to enhance the membrane contrast. Following several acetone washes, infiltration is carried out in the same cryotubes on an orbital shaker: 30% for 2–5 h, 70% for 3–6 h, and 90% overnight at room temperature. The next day, specimens are transferred into the caps of BEEM capsules containing 100% Epon/Araldite resin. The resin is replaced three times (2 h each). The specimens are cured in an oven (60◦C) for 48 h.

### Sectioning

For observation in a transmission electron microscope, specimens must be sliced thin so that electrons can penetrate through the tissue and generate an image. Ultrathin sections (40 nm) of specimens are cut using a diamond knife and collected on pioloform-coated single slot grids. For C. elegans, we orient the animal so that it is perpendicular to the sectioning surface (**Figure 1**, bottom). We trim the animal to the reflex of the gonad using a glass knife. We collect ribbons of 250 sections from each animal. For mouse hippocampal neurons, we first remove the sapphire disks from the sample by submerging them in the liquid nitrogen for a few seconds. About 40 sections are cut from the exposed surface (**Figure 1**, bottom) and collected. These sections are stained with 2.5% uranyl acetate in 70% methanol for 4 min prior to imaging.

### Imaging

A transmission electron microscope is operated at 80–120 keV. Images are acquired with a digital camera. Roughly 200–300 images are collected from each time point.

### Image Analysis

The morphological analysis is performed in ImageJ using a custom-written macro. This macro records X- and Y-coordinates of hand-traced membrane structures such as the plasma membrane, active zone, membrane invaginations, synaptic vesicles, dense-core vesicles, large vesicles, and endosomes. The positional information is first exported as text files and then imported into MATLAB (MathWorks). We have written scripts in MATLAB to analyze the coordinates and calculate several statistics: distribution of vesicles from the plasma membrane, distribution of vesicles from the active zone, diameter of vesicles, and number of vesicles. A normality test is performed on these data, and P-values are calculated using student's T-test for normally distributed data and Mann– Whitney U-test for skewed data. The confidence level is set at 95%. For multiple comparisons, the Bonferroni correction is applied.

## RESULTS

### Accuracy of Timing

To test the accuracy of the timing for light stimulation, we applied a single stimulus and froze tissues at desired time points. We calculated the actual time at which the specimens were frozen by monitoring the time when the pressure was applied to the specimens. The pressure application can be monitored using an accelerometer for EM PACT 2 (Watanabe et al., 2013b, 2014a). For HPM 100, we recorded the signal that triggers opening of the valve that allows pressurized liquid nitrogen to pass through because specimens are frozen exactly 80 ms after the valve opening (see Section Materials and Methods; Watanabe et al., 2013a). On the EM ICE, we can measure the interval between the end of the desired stimulation program and the time at which the sensor, which is located close to the specimen, reaches 0◦C. The sensor is placed before the specimen, and it requires additional 3-6 ms for the specimen to experience the same magnitude of cooling.

We recorded the values from each machine on at least three different experimental days. Out of 52 shots taken on the EM PACT2 (4 experiments, 13 shots/experiment), the desired time point was achieved in only 5 shots (**Figure 3**). The actual timing was variable, ranging from −102 to 110 ms. Because of this variability, the actual timing is always calculated post-hoc for each shot on the EM PACT2. On the HPM100, nearly two thirds of the shots (57/90) achieved the desired time point (3 experiments, 30 shots/experiment). The rest were either 5 ms before or after the desired time point, indicating that the variation is small in the HPM 100. However, the signal for the valve opening is measured at 5 ms time interval (200 Hz), and thus each recording has an uncertainty of ±2.5 ms. The EM ICE, on the other hand, records the temperature and pressure sensors' signals at 2000 Hz. The data from the EM ICE indicated that 56 out of 72 shots (3 experiments, 24 shots/experiment) landed between 1.5 and 2.5 ms after the desired time point. The maximum delay was 4 ms, and this maximum delay occurred on the very first shot on each experimental day. The rest of the shots were within 1.5–3.5 ms after the desired time point (average = 2.2 ± 0.08 ms). Given that the additional 3 ms is required to freeze specimens at 5 µm deep from the surface, a total of 4.5–6.5 ms is expected. These results indicate that millisecond temporal control can be achieved in HPM 100 and EM ICE.

### Exocytosis

To further test the accuracy of the timing, we stimulated C. elegans neuromuscular junctions for 20 ms and froze them at the end of the pulse using the EM PACT2. Based on the electrophysiological recordings, channelrhodopsin induced neurotransmission is close to the peak at this time point in the C. elegans motor neurons (Watanabe et al., 2013b). From a total of 13 shots taken, only 1 shot achieved this time point. To further test the accuracy, we also imaged the samples that were frozen 10 ms before the onset of the light pulse and 30 ms after the end of the light pulse, as calculated post-hoc. We were able to capture exocytic omega figures at neuromuscular junctions frozen 20 ms after the light onset (**Figure 4A**). However, these structures were not observed in the samples frozen 10 ms before the light onset or 30 ms after the end of the light pulse. These results indicate that the EM PACT2, although unreliable, can be used to capture fast membrane dynamics at synaptic terminals.

To test if the exocytic omega figures can be captured with the HPM 100 and the EM ICE, we froze mouse hippocampal neurons expressing channelrhodopsin. We applied a single 10 ms light pulse to the specimen and froze it 5 ms after the end of the light pulse. In the neurons frozen on the HPM 100, we found at least one omega figure in 19% of synapses at this time point (**Figure 4B**; 193 profiles; Watanabe et al., 2013a). Similarly, omega figures were captured in 18% of synapses by EM ICE (**Figure 4C**; 105 profiles). These results indicate that the flash-and-freeze approach can reliably capture the membrane dynamics at synapses with milliseconds temporal resolution.

### DISCUSSION

### Comparisons among Three Instruments

To visualize membrane dynamics at synapses, we developed a technique, "flash-and-freeze," which couples optogenetic stimulation of neurons with high-pressure freezing. Here, we

FIGURE 4 | Representative micrographs showing exocytic omega figures from the flash-and-freeze experiments. (A) *C. elegans* neuromuscular junctions, stimulated for 20 ms, and frozen at the end of the light pulse. Mouse hippocampal neurons, stimulated for 10 ms, frozen 5 ms after the end of the pulse using HPM 100 (B), and EM ICE (C). These structures are rarely observed in non-stimulated control or in specimens frozen at later time points after the stimulation.

compared three instruments that allow such experiments: EM PACT2, HPM100, and EM ICE. There are advantages and disadvantages for each instrument. The major advantage of the EM PACT 2 is in its potential application - higher intensity of ultraviolet light can be applied to specimens because light only has to penetrate through a clear sapphire anvil. Most caged compounds are uncaged with ultraviolet light. In both HPM100 and EM ICE, light must travel through an optical fiber and penetrate through a plastic sample holder that absorbs ultraviolet light. Thus, experiments using caged compound are more feasible in the EM PACT2. However, the EM PACT2 required many trials before the desired time point was reached. Besides the unreliability, the major disadvantage of the EM PACT2 is the size of the specimen cup. A specimen cup has a dimension of 1.6 mm diameter × 100 or 200 µm depth. Cells must be cultured on sapphire disks with a diameter of 1.4 mm. These small sapphire disks float in culture media, making it difficult to culture cells. On the other hand, sapphire disks as large as 6 mm can be loaded into the HPM100 or the EM ICE. Both the HPM100 and the EM ICE are essentially comparable in terms of the freezing quality and the temporal precision of light stimulation. However, HPM100 is more flexible in terms of its potential applications, as the light stimulation device and programs are customizable. With the expansion in the repertoire of optogenetic tools or lightsensitive compounds, cellular activity can be controlled using different wavelengths of light, and it is possible to develop a multi-color system for the HPM100. The major disadvantages of HPM 100 are that it requires an external device and that the actual freezing point must be calculated based on when the liquid nitrogen valve opens each time. On the contrary, EM ICE does not require a custom device or an external light source—everything is integrated. Furthermore, it achieves the most precise control of the light stimulation and the fastest freezing rate, requiring only 3 ms from the initial temperature drop to −20◦C. Despite these differences among instruments, exocytosis of synaptic vesicles, which occurs on a millisecond time scale, were captured in all three instruments, suggesting that any of these machines are compatible with flash-and-freeze experiments. Therefore, the choice should be made depending on the potential applications.

### Potential Applications

The flash-and-freeze approach can be applied to study many cell biological problems. We have been using channelrhodopsin to stimulate neuronal activity and capture membrane dynamics at synapses. Neuronal activity or cellular functions can be manipulated using light-sensitive molecules such as bacterial opsins (Han and Boyden, 2007; Zhang et al., 2007a; Berthold et al., 2008; Chow et al., 2010), caged compounds (Walker et al., 1986; Tsien et al., 1987; Milburn et al., 1989; Adams and Tsien, 1993; Wieboldt et al., 1994), light-sensitive proteins (Levskaya et al., 2009; Wu et al., 2009; Yazawa et al., 2009; Idevall-Hagren et al., 2012; Zhou et al., 2012; Guntas et al., 2015), and photoisomerizable molecule (Banghart et al., 2004; Kramer et al., 2009). These tools can be used to inhibit or activate neuronal activity, protein function, lipid composition, signaling cascades, and ion composition in vesicular structures. The morphological changes that take place during these manipulations can be preserved through the flash-and-freeze approach and visualized with a millisecond temporal resolution. Furthermore, there are neurons like photoreceptors that are naturally sensitive to light. The flash-and-freeze approach can be readily adapted to address the mechanism underlying tonic release of neurotransmitters from these terminals and how exocytosis is mediated by synaptic ribbons, electron dense proteinaceous protrusion in the center of the active zone (LoGiudice and Matthews, 2009).

### Potential Difficulties

Membrane dynamics are inferred by observing the membrane morphology at defined time points, but individual electron micrographs still capture static images of cells. The flash-andfreeze technique cannot follow the membrane dynamics in a single cell. Therefore, to reconstruct the membrane dynamics, hundreds of images must be acquired and analyzed from each time point. For membrane trafficking at synapses, an even higher number may be required due to the heterogeneity in the release probability (Rosenmund and Stevens, 1996)—not every action potential leads to fusion of synaptic vesicles. We have estimated that about 20–30% of synapses are activated with a single light pulse in mouse hippocampal neurons (Watanabe et al., 2013a, 2014b). To produce data with statistical significance, we have analyzed over 10,000 images. These analyses were performed blind to genotypes and time points, and thus the average frequency of morphological features at a particular time point can be determined from these images. To expedite analysis and increase the statistical power, automation is in an immediate need. However, with the recent advances in methods for automated image acquisition and analysis (Potter et al., 1999; Denk and Horstmann, 2004; Knott et al., 2008; Hayworth et al., 2014), achieving such a goal is likely within reach.

### AUTHOR CONTRIBUTIONS

SW performed all the experiments, analyzed the data, and wrote the manuscript.

## ACKNOWLEDGMENTS

I would like to thank Erik Jorgensen, Christian Rosenmund, and all the members in their laboratories for the help in performing the original experiments. The light stimulation device for Leica EM PACT 2 and HPM 100 is designed and developed by M. Wayne Davis. I would like to thank the Grass Foundation and the Marine Biological Laboratory at Woods Hole for space, equipment and funding for performing these experiments. I would like to thank Sumana Raychaudhuri and Edward Hujber for a critical reading of the manuscript. The research is supported by Johns Hopkins University startup fund.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnsyn. 2016.00024

### REFERENCES


properties, and activation of glutamate receptors on a microsecond time scale. Proc. Natl. Acad. Sci. U.S.A. 91, 8752–8756. doi: 10.1073/pnas.91.19.8752


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Watanabe. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Postsynaptic Density: There Is More than Meets the Eye

Ayse Dosemeci <sup>1</sup> \*, Richard J. Weinberg<sup>2</sup> , Thomas S. Reese<sup>1</sup> and Jung-Hwa Tao-Cheng<sup>3</sup>

<sup>1</sup> Laboratory of Neurobiology, National Institute of Neurological Disorders and Stroke (NINDS), National Institutes of Health (NIH), Bethesda, MD, USA, <sup>2</sup> Department of Cell Biology and Physiology, University of North Carolina, Chapel Hill, Chapel Hill, NC, USA, <sup>3</sup> Electron Microscopy Facility, National Institute of Neurological Disorders and Stroke (NINDS), National Institutes of Health (NIH), Bethesda, MD, USA

The postsynaptic density (PSD), apparent in electron micrographs as a dense lamina just beneath the postsynaptic membrane, includes a deeper layer, the "pallium", containing a scaffold of Shank and Homer proteins. Though poorly defined in traditionally prepared thin-section electron micrographs, the pallium becomes denser and more conspicuous during intense synaptic activity, due to the reversible addition of CaMKII and other proteins. In this Perspective article, we review the significance of CaMKII-mediated recruitment of proteins to the pallium with respect to both the trafficking of receptors and the remodeling of spine shape that follow synaptic stimulation. We suggest that the level and duration of CaMKII translocation and activation in the pallium will shape activity-induced changes in the spine.

#### Keywords: postsynaptic density, PSD, pallium, electron microscopy, EM, Shank, Homer, CaMKII

Excitatory glutamatergic synapses examined with the electron microscope typically display a pronounced postsynaptic density (PSD), which appears in conventional electron micrographs as an approximately 30 nm thick electron-dense structure applied to the postsynaptic membrane (Palay, 1956). The large majority of excitatory synapses in the vertebrate CNS release glutamate as neurotransmitter. Ionotropic glutamate receptors concentrate at the PSD, where specialized molecules anchor them and regulate their trafficking; modulation of their expression and trafficking plays a key role in synaptic plasticity. Decades of research show that much of the story of this modulation resides in the PSD.

Using special stains, Valtschanoff and Weinberg (2001) uncovered a distinct layer ∼50 nm thick immediately adjacent and deep to the PSD. Though previously reported, this ''subsynaptic web'' (DeRobertis, 1964) had been largely ignored for decades, because it is difficult to distinguish from the rest of the spine cytoplasm after standard histological procedures. Extending this work, Petralia et al. (2005) reported high levels of immunoEM label for both Shank and Homer in the same region, which was referred to as the ''subjacent'' area.

The subsynaptic web becomes more prominent after short bouts of stimulation, reflecting the reversible incorporation of additional proteins. Because this part of the PSD comprises different proteins and displays a more labile organization than the ''core'' layer lining the synaptic membrane, it merits a distinct name; we propose to designate this separate structural and functional layer of the PSD as its pallium or mantle (**Figure 1**). Here we present a perspective on PSD structure that highlights potential functions carried out in the pallium.

### CORE LAYER OF THE PSD

Though highly variable, PSDs in the rodent forebrain contain on average 300–400 copies of PSD-95 and related membrane-associated guanylate kinase (MAGUK) molecules (Chen et al., 2005;

#### Edited by:

George Augustine, Nanyang Technologicial University (NTU), Singapore

#### Reviewed by:

Eric Hanse, University of Gothenburg, Sweden Anne McKinney, McGill University, Canada

### \*Correspondence:

Ayse Dosemeci dosemeca@mail.nih.gov

Received: 17 May 2016 Accepted: 25 July 2016 Published: 19 August 2016

#### Citation:

Dosemeci A, Weinberg RJ, Reese TS and Tao-Cheng J-H (2016) The Postsynaptic Density: There Is More than Meets the Eye. Front. Synaptic Neurosci. 8:23. doi: 10.3389/fnsyn.2016.00023

Scale bar: 0.1 µm.

Sugiyama et al., 2005). PSD-95 is localized in the core layer of the PSD. Comparing immunogold labeling with antibodies against its N- and C-termini shows that PSD-95 is oriented with its N-terminus near the plane of the postsynaptic membrane and its C-terminus deep in the spine (Chen et al., 2008). Electron microscopic tomography reveals vertical filaments extending from the postsynaptic membrane in numbers and dimensions expected for PSD-95 (Chen et al., 2008), and these filaments are depleted by acute knockdown of PSD-95 (Chen et al., 2011). Thus, PSD-95-containing filaments oriented vertically to the postsynaptic membrane represent a major structural component of the core layer (Chen et al., 2008). This close apposition to the membrane puts PSD-95 in a position to bind neurotransmitter receptors. Indeed, there is substantial evidence indicating the association of PSD-95 with both NMDA and AMPA types of glutamate receptors.

deeper from the PSD core is a specialized region we call the PSD pallium.

EM tomography also reveals filaments of different lengths, oriented parallel to the postsynaptic membrane, that appear to contact the deep (C-terminal) ends of PSD-95 filaments (Chen et al., 2008). The dimensions of one frequently-encountered type of horizontal filament corresponds to that expected for guanylate kinase-associated protein (GKAP), consistent with biochemical work demonstrating that GKAP can bind PSD-95 near its C-terminal (Kim et al., 1997). GKAP is in the PSD at a ratio of approximately one GKAP for two PSD-95s (Lowenthal et al., 2015) so a layer of GKAP cross-linked with PSD-95 could provide an interface between the core layer and the pallium of the PSD. Like PSD-95 (Yang et al., 2011), GKAP stays put during brief stimulation (Tao-Cheng et al.,

### BOX 1 | Major scaffold proteins in the PSD.

Names used in the present article are listed along with synonyms.

MAGUKs (membrane-associated guanylate kinases).

The most abundant MAGUK at the mammalian PSD is PSD-95 (DLG4, SAP90); others include SAP97 (DLG1), PSD-93 (DLG2, Chapsyn-110), and SAP102 (DLG3).

GKAP (guanylate kinase-associated protein) also called SAPAP (SAP90/PSD-95-associated protein), or disks large-associated protein.

Shank (SH3 and multiple ankyrin repeat domains protein) also called ProSAP (Proline-rich synapse-associated protein), Synamon and CortBP (Cortactin binding protein).

Homer also called Vesl, Cupidin, and PSD-Zip45.

2015). Thus, the dense matrix of PSD-95/MAGUK filaments capped by GKAPs provides a stable platform for the subjacent pallium.

## PALLIAL LAYER OF THE PSD

Early biochemical analyses indicated that the PSD complex contains several types of scaffolding molecules in addition to MAGUKs and GKAPs (**Box 1**). Two-hybrid and co-immunoprecipitation experiments led to the identification of the Shank/ProSAP family of proteins, which bind to GKAPs (Boeckers et al., 1999a,b; Naisbitt et al., 1999). Similar twohybrid studies identified the Homer family of proteins as binding partners for Shanks and the two types of proteins were shown to co-immunoprecipitate from brain extracts (Tu et al., 1999). Both Shank and Homer family proteins are highly enriched in PSD fractions (Xiao et al., 1998; Naisbitt et al., 1999) and co-purify as large complexes, with PSD-95, GKAP and other PSD constituents (Collins et al., 2006; Dosemeci et al., 2007).

While biochemical studies identify Shank and Homer as part of the PSD complex, immunoEM studies show that label for these proteins concentrates outside the electron-dense material conventionally defined as the PSD (**Figure 2**). Although the reported distributions of label for these proteins differ somewhat between groups, immunoEM studies typically find that much of the postsynaptic immunogold label for both Shank and Homer lie in a layer immediately below the PSD core (Tu et al., 1999; Valtschanoff and Weinberg, 2001; Petralia et al., 2005; Rostaing et al., 2006; Tao-Cheng et al., 2010, 2014a).

These data imply that a network containing Shanks and Homer extends the PSD scaffold well beyond the classically recognized electron-dense material. This second layer, which we term the pallium, is likely pegged to the PSD core through GKAPs, which can simultaneously associate with PSD-95 and Shanks. In vitro studies report that Shanks can associate with each other through their SAM domains to form sheet-like structures (Baron et al., 2006) and Homers can polymerize into tetramers that can cross-link Shanks (Hayashi et al., 2009). Furthermore, purified C-terminal Shank and Homer form a mesh-like matrix when mixed together (Hayashi et al., 2009). Thus, Shank and Homer proteins in the pallium are likely to form an extended scaffold. The side of this scaffold facing the cleft is continuous

with the PSD core, whereas the cytoplasmic side, characterized by an extremely rough surface (Petersen et al., 2003) merges imperceptibly into the central zone of the spine head.

### ACTIVITY-INDUCED CHANGES AT THE PALLIUM

### The Pallium Becomes More Electron-Dense as CaMKII Accumulates

Thin-section EM of cultured hippocampal neurons reveals increased electron density at the pallium under excitatory conditions (**Figure 3**: top panels), a phenomenon we have described as PSD ''thickening'' (Dosemeci et al., 2001). The increase in electron density likely reflects the addition of proteins, and immunoEM shows significant increase in CaMKII immunolabel within the PSD complex under excitatory conditions (**Figure 3**: bottom panels, Dosemeci et al., 2001). Similar morphological changes at the PSD, accompanied by the accumulation of CaMKII and other proteins, also occur under the excitotoxic conditions promoted by ischemia (Suzuki et al., 1994; Hu et al., 1998; Martone et al., 1999). The duration of the morphological and compositional changes at the PSD appears to vary according to the type of stimulation; importantly, more sustained modification is observed following an LTP-inducing protocol (Otmakhov et al., 2004).

### Role of CaMKII as a Kinase

CaMKII, a Ca2+-regulated serine-threonine protein kinase, is present at very high concentrations in neurons, with a relative abundance reaching ∼1–2% of total protein in the cerebral cortex and hippocampus (Erondu and Kennedy, 1985). Within neurons, CaMKII is especially prominent in the PSD pallium (Petralia et al., 2005; Ding et al., 2013). Activation of CaMKII plays a pivotal role in certain types of synaptic modification, most notably NMDA-dependent LTP (reviews: Lisman et al., 2012; Shonesy et al., 2014).

The CaMKII holoenzyme is a dodecamer made of individually active subunits, with α- and β-isoforms prevalent in brain. Subunits within a single holoenzyme can phosphorylate one another, a process called autophosphorylation. Upon autophosphorylation in the presence of Ca2+, CaMKII becomes autonomous, allowing enzymatic activity to persist beyond the duration of Ca2<sup>+</sup> elevation (reviews: Hell, 2014; Shonesy et al., 2014). Autonomous CaMKII can remain at the PSD pallium after the cessation of excitatory stimuli (Dosemeci et al., 2002; Otmakhov et al., 2004). Continued co-localization of autonomous CaMKII with its substrates at the pallium would counteract dephosphorylation by phosphatases and thus help maintain phosphorylation of CaMKII substrates.

The accumulation and activation of CaMKII at the pallium triggers further changes at the PSD. Some of these have been studied in depth, such as the incorporation of AMPA receptors

(right). Upon intense synaptic activity the pallium of the PSD becomes electron-dense due to the accumulation of proteins, including CaMKII (bottom panels). Scale bar: 0.1 µm.

into the postsynaptic membrane (review: Lisman et al., 2012). ImmunoEM studies also reveal activity-induced increases in the levels of a number of proteins at the pallium, including SynGAP, AIDA-1 and Shanks, in parallel with CaMKII. Importantly, the redistribution of all these PSD components is blocked by the application of a specific CaMKII inhibitor (Yang et al., 2013; Tao-Cheng et al., 2014b; Dosemeci et al., 2016), indicating that the activation of CaMKII is required for the redistribution of these PSD components.

### Role of CaMKII as a Dynamic Structural Element

In addition to its function as a protein kinase, accumulating evidence suggests that CaMKII in the pallium may act as a scaffold. Bingol et al. (2010) show that activity-induced translocation of autophosphorylated α-CaMKII, but not its kinase activity, is responsible for the translocation of proteasomes into synapses. Similarly, activity-induced CaMKII translocation is responsible for parallel accumulation of the deubiquitinase CYLD, specific for K63-linked polyubiquitins, to the PSD (Thein et al., 2014). Since K63 linked polyubiquitination inhibits interaction of proteins with proteasomes (Nathan et al., 2013) CYLD activity is expected to facilitate protein binding to proteasomes. Thus, CaMKII accumulated at the pallium appears to constitute a structural platform that brings together components that may work synergistically to promote local degradation of proteins.

Activity-induced translocation of CaMKII also may mediate modifications in spine morphology. CaMKII holoenzymes at the spine cytoplasm serve as actin bundling elements (Okamoto et al., 2007) and dissociation of CaMKII from F-actin upon autophosphorylation destabilizes the actin cytoskeleton (Kim et al., 2015). Thus the translocation of CaMKII from the actin network to the PSD pallium can trigger synchronized changes at the PSD and actin cytoskeleton.

### SynGAP Levels Decrease in the PSD Core and Increase in the Pallium

SynGAP is an enzymatic regulator of Ras, a small GTPase. However, as one of the most abundant proteins at the PSD (at levels exceeding even those of PSD-95, its binding partner at the PSD core), SynGAP is likely to also play a non-enzymatic role. ImmunoEM shows that SynGAP label is concentrated at the PSD core (Yang et al., 2011, 2013). After strong depolarization, labeling for SynGAP significantly decreases at the PSD core while increasing at the pallium (Yang et al., 2011, 2013). Under the same conditions, label for PSD-95 does not change its localization at the PSD (Yang et al., 2011), implying dissociation of the two molecules. Further studies indicate that the release of SynGAP from PSD-95 and its exit from the PSD core require CaMKIImediated phosphorylation (Yang et al., 2013; Araki et al., 2015).

NMDA- or high K+-induced changes in the distribution of SynGAP are reversed within 30 min after cessation of the excitatory conditions. In contrast, a stimulation protocol (glycine in the absence of Mg2+) that leads to sustained increases in synaptic efficacy (chem-LTP) also leads to a sustained exclusion of SynGAP from the spine (Araki et al., 2015) indicating a correlation between LTP and the removal of SynGAP. It appears likely that a wide range of excitatory stimuli promotes translocation of SynGAP out of the PSD core, but that only under LTP-promoting conditions is the molecule removed on a long-term basis. The extent and maintenance of SynGAP phosphorylation at the pallium may determine its subsequent movement. In this regard, it is interesting that SynGAP is most efficiently phosphorylated by the autonomous form of CaMKII in the absence of Ca2<sup>+</sup> (Dosemeci and Jaffe, 2010).

What might be the functional implications of activity-induced redistribution of SynGAP at the PSD? Both SynGAP and members of the transmembrane AMPA receptor regulatory protein family (TARPs) can bind to the same PDZ domains on PSD-95 (Kim et al., 1998; Schnell et al., 2002; Dakoji et al., 2003) and thus may compete for association with PSD-95. Considering its high abundance in the PSD, SynGAP is likely to block TARP binding to PDZ domains on PSD-95 under basal conditions. Thus, the removal of SynGAP during activity may be a prerequisite for the anchoring of AMPA receptors to the PSD, explaining the well-documented role of SynGAP as an inhibitor of AMPA receptor insertion (Rumbaugh et al., 2006). This function may account for the remarkable abundance of SynGAP at the PSD.

### AIDA-1 Levels Decrease in the PSD Core and Increase in the Pallium

Quantitative mass spectrometric analysis of PSD fractions reveals that Amyloid-β protein precursor Intracellular DomainAssociated 1 (AIDA-1, also known as EB-1 and ANKS1B) is a major component of the PSD complex, with a PSD-95/AIDA-1/GKAP stoichiometry of 2:1:1 (Lowenthal et al., 2015). Like SynGAP, AIDA-1 binds directly to PSD-95 at the same domains that bind TARPs (Jordan et al., 2007) and therefore could also interfere with the anchoring of AMPA receptors. By immunoEM, the AIDA-1 label is mostly located within the PSD core at rest (Jacob et al., 2010; Dosemeci et al., 2015). Under excitatory conditions AIDA-1 label at the PSD core is significantly reduced with a concomitant increase at the pallium (Dosemeci et al., 2015). The reversible CaMKII-mediated redistribution of AIDA-1 at the PSD under excitatory conditions (Dosemeci et al., 2016) parallels that of SynGAP. We speculate that different regulatory mechanisms may trigger dissociation of SynGAP and AIDA-1 from PSD-95.

### Shank Levels Increase in the Pallium

Shanks, a protein scaffold family concentrated in the PSD, are encoded by three genes, Shank1, Shank2 and Shank3, each giving rise to multiple splice variants. Shank mutations have been linked to autism and other neurodevelopmental/neuropsychiatric disorders (reviews: Grabrucker et al., 2011; Sala et al., 2015). All Shank isoforms have a similar molecular organization, with specialized domains that can associate with GKAPs, Homers, and other Shanks. Analysis of immunoEM data suggests that Shanks at the PSD are organized into a proximal pool, close enough to the interface between the core and pallium to associate with GKAP, and a distal (deeper) pool that may be stabilized through association with Homers and/or with other Shanks. Under excitatory conditions, Shanks accumulate preferentially in the distal pool (Tao-Cheng et al., 2015).

Shanks promote maturation of dendritic spines and the enlargement of spine heads (Sala et al., 2001), although the precise mechanism remains unclear. Considering that changes in spine shape and size involve reorganization of the actin cytoskeleton, it is likely that Shanks regulate spine morphology through interaction with actin. Indeed, Shanks associate with three actin-regulating proteins, Insulin Receptor Substrate Protein 53 (IRSp53), Abp1 and cortactin.

The actin binding protein IRSp53 (also called BAIAP2; see review by Kang et al., 2016) is a major PSD component, with a molar ratio of IRSp53 to PSD-95 of 1:4 (Lowenthal et al., 2015). IRSp53 is located between layers containing Shank and PSD-95, relatively close to the postsynaptic membrane (Burette et al., 2014), where it may function as a linker between the actin cytoskeleton and the plasma membrane (Scita et al., 2008; Burette et al., 2014). Indeed, IRSp53 contains an actin-binding BAR-like domain that can induce changes in membrane curvature (Zhao et al., 2011) and thus may be involved in activity-induced changes in the curvature of the synapse (Burette et al., 2014).

Abp1, which can associate simultaneously with actin and Shanks, may link the actin cytoskeleton to the PSD (Qualmann et al., 2004). Abp1 preferentially interacts with dynamic rather than static F-actin structures (Kessels et al., 2000) and more Abp1 becomes incorporated into Shank-positive synapses during activity (Qualmann et al., 2004). Overexpression of Abp1

#### TABLE 1 | Redistribution of postsynaptic density (PSD) components during activity.


Changes observed in immunogold label density for selected proteins following treatment of cultured hippocampal neurons with media containing NMDA or high K<sup>+</sup> .

increases the density of mushroom-shaped spines; importantly, its association with Shank is necessary for Abp1-mediated regulation of spine morphology (Haeckel et al., 2008).

Another protein that associates with both Shanks and actin is cortactin, although immunoEM shows that cortactin concentrates mainly at the central region of the spine (Racz and Weinberg, 2004). In contrast to Abp1, cortactin exits the spines during activity (Hering and Sheng, 2003). While Abp1 overexpression increases mushroom spine density (Haeckel et al., 2008), cortactin overexpression causes elongation of spines (Hering and Sheng, 2003).

The above considerations lead us to suggest a model for activity-induced modification of spine morphology: under basal conditions, cytosolic Shanks within the spine are pegged to the actin cytoskeleton through cortactin and held outside of the pallium. During activity, CaMKII-mediated phosphorylation releases Shanks from cortactin. The Shank thus released accumulates at the pallium, while cortactin exits the spine. Shank accumulated at the pallium could then recruit Abp1 to promote remodeling of the actin cytoskeleton around the pallium.

### ACTIN AND THE PSD

There are several reports that filaments of F-actin, the primary cytoskeletal component of the spine, contact the PSD (Gulley and Reese, 1981; Landis and Reese, 1983; Fifková, 1985), especially at its periphery (Burette et al., 2012), though these contacts are likely highly variable, considering the dynamic nature of F-actin. While the molecular basis of the attachment of actin filaments to the PSD remains uncertain, a number of PSD molecules are plausible candidates. As discussed above, two Shank binding proteins, IRSp53 and Abp1, may provide a link between the PSD and the actin cytoskeleton.

Yet another point of interaction of actin with the PSD may be through the ubiquitous actin binding protein α-actinin. In vitro assays show that α-actinin can form a ternary complex with the PSD proteins, densin and CaMKII (Walikonis et al., 2001). Loss of α-actinin-2 in rat hippocampal neurons creates an increased density of immature, filopodia-like protrusions that fail to mature into a mushroom-shaped spine during development (Hodges et al., 2014).

### CONCLUSION

The pallium is an extension of the electron-dense PSD core, demarcated by strong immunolabeling for two PSD Dosemeci et al. The Postsynaptic Density

scaffold proteins, Shank and Homer. This specialized zone, sandwiched between the electron-dense PSD core and the actin ''spinoskeleton'', is highly dynamic, exhibiting striking changes during synaptic activity. Under conditions of strong excitation, the pallium becomes electron-dense, with the addition of CaMKII and several other proteins (**Table 1**). Activation and/or translocation of CaMKII is necessary for the recruitment of other components to the pallium.

The pallium can be viewed as a hub where several proteins converge during activity. Accumulating evidence on the movements of individual proteins suggests a mechanism for concerted insertion of receptors to the PSD and re-organization of the actin spinoskeleton, both mediated by CaMKII. Upon synaptic stimulation, translocation and activation of CaMKII cause SynGAP and AIDA-1 to move out of the PSD core and accumulate at the pallium (Yang et al., 2013; Dosemeci et al., 2015, 2016). We propose that the movement of SynGAP (and perhaps also of AIDA-1) vacates ''slots'' on PSD-95, providing a window of opportunity for the insertion of AMPA receptors. Simultaneous CaMKII-mediated accumulation of Shanks at the pallium (Tao-Cheng et al., 2014b), on the other hand, may enable actin reorganization around the PSD. CaMKII also acts as a dynamic structural element whose activity-induced translocation changes the molecular organization within the spine. Dissociation of CaMKII from F-actin causes destabilization and reorganization

### REFERENCES


of the actin cytoskeleton (Kim et al., 2015). Subsequent accumulation of CaMKII at the pallium docks elements that regulate protein turnover (Bingol et al., 2010; Thein et al., 2014).

In summary, activation of CaMKII and its translocation to the pallium under excitatory conditions trigger a chain of events poised to elicit profound effects on the organization of the PSD complex and of the actin cytoskeleton that could synchronize receptor trafficking with changes in spine morphology. We speculate that the degree and duration of CaMKII accumulation and activity at the pallium, promoted by different types of excitatory stimuli, may determine the type and level of synaptic modification.

### AUTHOR CONTRIBUTIONS

All authors contributed to the writing of the manuscript.

### ACKNOWLEDGMENTS

We would like to thank Dr. John Lisman for stimulating discussions and for a critical reading of the manuscript. This work was supported by the Intramural Research Program of the National Institute of Neurological Disorders and Stroke (NINDS)/National Institutes of Health (NIH) and NINDS/NIH NS 039444 grant to RJW.


of Ca2+/calmodulin-dependent protein kinase II and α-actinin. J. Neurosci. 21, 423–433.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Dosemeci, Weinberg, Reese and Tao-Cheng. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Optogenetic Monitoring of Synaptic Activity with Genetically Encoded Voltage Indicators

Ryuichi Nakajima1† , Arong Jung1,2† , Bong-June Yoon<sup>2</sup> and Bradley J. Baker 1,3 \*

<sup>1</sup> Center for Functional Connectomics, Korea Institute of Science and Technology, Seongbuk-gu, Seoul, South Korea, <sup>2</sup> College of Life Sciences and Biotechnology, Korea University, Seongbuk-gu, Seoul, South Korea, <sup>3</sup> Department of Neuroscience, Korea University of Science and Technology, Daejeon, South Korea

The age of genetically encoded voltage indicators (GEVIs) has matured to the point that changes in membrane potential can now be observed optically in vivo. Improving the signal size and speed of these voltage sensors has been the primary driving forces during this maturation process. As a result, there is a wide range of probes using different voltage detecting mechanisms and fluorescent reporters. As the use of these probes transitions from optically reporting membrane potential in single, cultured cells to imaging populations of cells in slice and/or in vivo, a new challenge emerges—optically resolving the different types of neuronal activity. While improvements in speed and signal size are still needed, optimizing the voltage range and the subcellular expression (i.e., soma only) of the probe are becoming more important. In this review, we will examine the ability of recently developed probes to report synaptic activity in slice and in vivo. The voltagesensing fluorescent protein (VSFP) family of voltage sensors, ArcLight, ASAP-1, and the rhodopsin family of probes are all good at reporting changes in membrane potential, but all have difficulty distinguishing subthreshold depolarizations from action potentials and detecting neuronal inhibition when imaging populations of cells. Finally, we will offer a few possible ways to improve the optical resolution of the various types of neuronal activities.

#### Edited by:

George Augustine, Nanyang Technological University, Singapore

#### Reviewed by:

Jason D. Shepherd, University of Utah, USA Christian Wilms, Scientifica Ltd., UK

### \*Correspondence:

Bradley J. Baker bradley.baker19@gmail.com

†These authors have contributed equally to this work.

> Received: 06 May 2016 Accepted: 25 July 2016 Published: 05 August 2016

#### Citation:

Nakajima R, Jung A, Yoon B-J and Baker BJ (2016) Optogenetic Monitoring of Synaptic Activity with Genetically Encoded Voltage Indicators. Front. Synaptic Neurosci. 8:22. doi: 10.3389/fnsyn.2016.00022 Keywords: genetically-encoded voltage indicators, synaptic activity, optogenetics, brain slices, in vivo

### INTRODUCTION

As developers of genetically-encoded voltage indicators (GEVIs) we are often asked for our best probe. Until recently, a good GEVI would have been any that gave a voltage-dependent, optical signal in mammalian cells (Dimitrov et al., 2007; Lundby et al., 2008, 2010; Perron et al., 2009a,b). Now the experimenter has several probes to choose from that differ in their voltage-dependencies, speed, signal size, and brightness (Akemann et al., 2012; Jin et al., 2012; Kralj et al., 2012; Han et al., 2013; St-Pierre et al., 2014; Zou et al., 2014; Gong et al., 2015; Piao et al., 2015; Abdelfattah et al., 2016). The combinations of these varying characteristics result in strengths and weaknesses of every GEVI available. There is no perfect probe that can optically resolve action potentials, synaptic activity, and neuronal inhibition in vivo. Some GEVIs will give large, voltage-dependent optical signals but are very dim limiting their usefulness in vivo. Others will give large optical signals but are very slow reducing their ability to resolve fast firing action potentials. So now, when asked which is the best probe, the answer is simply another question. What do you want to measure? To fit with the theme of this edition, we will assume that the answer to that question is synaptic activity.

Several reviews have been published comparing the signal size, speed, and brightness of the GEVIs currently available at the time of publication (Wachowiak and Knöpfel, 2009; Akemann et al., 2012, 2015; Knöpfel, 2012; Mutoh et al., 2012; Perron et al., 2012; Mutoh and Knöpfel, 2013; Emiliani et al., 2015; Knöpfel et al., 2015; St-Pierre et al., 2015; Storace et al., 2015b, 2016; Antic et al., 2016). In this review, we will shift the focus to one of the lesser considered characteristics of a GEVI, the voltage-sensitivity of the probe. Of course, the other characteristics, especially signal size and brightness, are still important, but the range and steepness of the voltage sensitivity of the optical response have extremely important consequences on which type of neuronal activity a GEVI reports well. For instance, a GEVI with a voltage range from −20 mV to +30 mV would be perfect for monitoring action potentials but not ideal for observing synaptic potentials. Even the shape of the slope of the optical response over the voltage range of the probe will affect its performance. The consequences of the slope and voltage range should also be considered when choosing probes for monitoring neuronal activity.

### A BRIEF DESCRIPTION OF CURRENTLY AVAILABLE GEVIs

There now exist several GEVIs with multiple mechanisms of converting membrane potential changes into an optical signal. These GEVIs fall into two main classes. One class utilizes bacterial rhodopsin to detect alterations in voltage, while the other class relies on a voltage-sensing domain (VSD) from voltage-sensing proteins. Another viable alternative for optical, neuronal recordings is hybrid voltage sensor (hVOS) which consists of a genetically encoded component, a farnesylated fluorescent protein (FP), and a quenching compound, dipicrylamine (DPA; Chanda et al., 2005; Wang et al., 2010, 2012; Ghitani et al., 2015). The requirement for the treatment with an exogenous chemical limits hVOS use in vivo but still has value for imaging voltage in slice preparations.

The molecular schematics of representative probes from these classes and their corresponding voltage ranges are shown in **Figure 1**. The voltage range of a generic mammalian neuron is color coded to represent different neuronal activities. The inhibitory postsynaptic potential (IPSP) voltage range is shown in blue. The excitatory postsynaptic potential (EPSP) voltage range is shown in yellow. Voltages corresponding to action potentials are color coded red. As can be seen from **Figure 1**, the slopes of these optical voltage responses are significantly different. This is an important consideration when measuring synaptic potentials. For instance, Butterfly 1.2 has nearly reached its maximal fluorescent change at −40 mV which would imply that differentiating subthreshold potentials from action potentials will be very difficult.

### Class I—The Rhodopsin-Based Probes

Channel rhodopsin has revolutionized neuroscience. The rhodopsin-based voltage sensors are promising to do the same

potential (IPSP), excitatory postsynaptic potential (EPSP) and action potential are indicated as blue, yellow and red, respectively. The vertical scale bar with minus ∆F/F indicates that the fluorescence dims upon depolarization of the plasma membrane. The voltage-sensitivity curves were as reported in: Arch (D95N; modified with permision from Kralj et al., 2012, Figures 3B, 5); Ace2N-mNeon (modified with permission from Gong et al., 2015, Figure 1D); ASAP-1 (modified with permission from St-Pierre et al., 2014, Figure 1D); ArcLight (modified with permission from Jin et al., 2012, Figure 1C); Butterfly 1.2 (modified with permission from Akemann et al., 2012, Figure 2C); hybrid voltage sensor (hVOS; modified with permission from Chanda et al., 2005, Figure 1D).

thing for imaging membrane potential. First developed in Adam Cohen's lab, the intrinsic fluorescence of rhodopsin as a Schiff base being protonated or deprotonated in response to voltage was used to image changes in membrane potential (Maclaurin et al., 2013). This probe, Arch, was extremely fast having a tau under 1 ms. The fast optical response is due in part to the fact that the chromophore resides in the voltage field enabling a nearly instantaneous response. The signal size was also large giving roughly a 70% ∆F/F optical signal per 100 mV membrane depolarization (Kralj et al., 2012; **Figure 1**, Arch (D95N)).

Arch excelled in speed and signal size but suffered from some serious weaknesses. The first weakness was that the original version had an associated, light-induced current. The D95N mutation drastically reduced this current but also resulted in a slower probe (Kralj et al., 2012). The second weakness was that it does not traffic well to the plasma membrane. Even with the addition of endoplasmic reticulum and Golgi network release motifs, every image of a rhodopsin probe in the literature exhibits high intracellular fluorescence (Kralj et al., 2012; Flytzanis et al., 2014; Gong et al., 2014, 2015; Hochbaum et al., 2014; Hou et al., 2014). The third and most devastating weakness was that Arch is very dim. The best versions of Arch and related probes are still at least 5× dimmer than the green fluorescent protein (GFP) requiring exceptionally strong illumination, at least 700× the light intensity required for ASAP-1 to visualize the probe activity (Flytzanis et al., 2014; St-Pierre et al., 2014).

The weak fluorescence of Arch limits its use to single cell in culture studies or to C. elegans (Kralj et al., 2012; Flytzanis et al., 2014) for two main reasons. The first is that the intrinsic fluorescence of higher order neuro-systems will mask the fluorescence of Arch-type probes. The second is that ∆F in addition to ∆F/F is an important characteristic of the GEVI when it comes to the signal to noise ratio. An example of this is shown in **Figure 2**. The HEK cell in **Figure 2** is expressing a GEVI from which the ∆F and the ∆F/F traces from three different light levels are shown (Lee et al., 2016). As can be seen from this comparison, a high ∆F/F value can be achieved by a large change in fluorescence or a small change in fluorescence when the probe is dim. Notice the increased noise in trace 3, a telltale sign of poor expression/dim fluorescence.

An ingenious solution to compensate for the poor fluorescence of the rhodopsin voltage probes was developed

showing averaged ∆F/F values from the same regions in (B; modified with permission from Lee et al., 2016, Figure 6).

simultaneously by the Adam Cohen and Mark Schnitzer laboratories. By fusing an FP to the rhodopsin protein, förster resonance energy transfer (FRET) enabled the rhodopsin chromophore to affect the fluorescence of the fused FP. This design reduced the excitation light intensity needed to visualize the GEVI while maintaining the speed of the optical response since the voltage-sensing chromophore was still in the voltage field. These probes could also cover different wavelengths since many different FPs could be fused to rhodopsin and give a signal (Gong et al., 2014; Zou et al., 2014). While this made the rhodopsin probes better, the optical signal sometimes could only indirectly report neuronal activity by determining the frequency of the noise in the optical recording (See Supplementary Figure 5 in Gong et al., 2014). Now, an exciting new version using the FP, mNeonGreen, has recently been reported (Gong et al., 2015). mNeonGreen is a very bright FP (Shaner et al., 2013) enabling Ace2N-mNeon to resolve action potentials in vivo in both flies and mice.

### Class II—VSD Containing GEVIs

The second class of GEVIs is also the oldest. The original GEVI, Flash (Siegel and Isacoff, 1997), was the result of inserting GFP downstream of the pore domain of the voltage-gated potassium channel, Shaker. Like the rhodopsin-based probes, the first generation of VSD-based probes had significant drawbacks making them useless in mammalian cells (Baker et al., 2007). The main problem was that the GEVIs did not traffic to the plasma membrane. In 2007, one of the biggest advancements in GEVI development was achieved by the Knöpfel laboratory when they fused FPs to the VSD of the voltage-sensing phosphatase gene from Ciona intestinalis (Murata et al., 2005). This probe, voltage-sensing fluorescent protein (VSFP) 2.1 trafficked well to the plasma membrane which resulted in the first voltagedependent optical signals from cultured neurons (Dimitrov et al., 2007).

Another issue with VSD-based GEVIs is that the chromophore resides outside of the voltage field so the optical signal relies on the conformational change of the VSD. These probes are therefore generally slower than the rhodopsin-based probes, but a recently developed red-shifted GEVI is extremely fast having taus under 1 ms (Abdelfattah et al., 2016).

There are three different designs for GEVIs that utilize a VSD. The first design uses a FRET pair flanking the VSD. An example is Butterfly 1.2 (Akemann et al., 2012). This probe is somewhat slow and gives a very small optical signal, less than 3% ∆F/F per 100 mV depolarization. A butterfly style probe that gives a faster and larger optical signal was developed last year called Nabi (Sung et al., 2015). An advantage of FRET-based probes is that the ratiometric imaging can remove movement artifacts due to respiration and blood flow in vivo. Theoretically, a ratiometric measurement could also be used to determine the absolute value of the membrane potential since the ratio is concentration independent. In practice, however, the relative fluorescence of the two chromophores differ substantially resulting in a potential increase in the noise for the analysis of the optical signal. Often the experimenter should only analyze the brighter signal (Wilt et al., 2013). It is also difficult to only excite the donor chromophore and not the acceptor as well. These factors combined with the relatively low signal size of FRET-based probes prohibit any reliable absolute measurement of membrane potential.

The second design involves a circularly-permuted fluorescent protein (cpFP) attached to the VSD. Initial designs fused the cpFP downstream of the VSD so that the chromophore was in the cytoplasm (Gautam et al., 2009; Barnett et al., 2012). Electrik PK gave very small signals less than 1% ∆F/F per 100 mV depolarization but were very fast having a tau under 2 ms. A substantial increase in signal size was achieved when the cpFP was placed between the S3 transmembrane segment and the S4 transmembrane segment of the VSD putting the chromophore outside of the cell (St-Pierre et al., 2014). This probe, ASAP-1, is one of the better GEVIs giving a fast and robust optical signal (tau = 1–2 ms and about 20% ∆F/F per 100 mV depolarization in HEK cells). ASAP-1 has a very broad voltage range which is virtually linear over much of the physiologically relevant potentials of neurons.

The third design of GEVIs that utilize a VSD simply fuses the FP at the carboxy-terminus which puts the chromophore in the cytoplasm. During a systematic test of different FPs fused at different linker lengths from the VSD done in collaboration by Vincent Pieribone's lab and Larry Cohen's lab, a point mutation on the outside of the FP, Super Ecliptic pHlorin (Miesenbock et al., 1998; Ng et al., 2002) converted an alanine to an aspartic acid improving the optical signal 15 fold from 1% ∆F/F to 15% per 100 mV depolarization of the plasma membrane (Jin et al., 2012). This negative charge on the outside of the β-can seems to affect the fluorescence of a neighboring chromophore when S4 moves since mutations that favor the monomeric form of the FP reduce the voltage-dependent optical signal substantially (Kang and Baker, 2016). Further development of ArcLight has gotten signals as high as 40% ∆F/F per 100 mV depolarization step (Han et al., 2013). While ArcLight has the drawback of being slow, its brightness and signal size make it one of the better probes for imaging in vivo and in slice. In 2015, two publications improving the speed of this sensor were published. One dramatically improved the off rate called Arclightening but reduced the signal size to under 10% ∆F/F per 100 mV depolarization (Treger et al., 2015). The other, Bongwoori, improved the speed of the sensor and shifted the voltage response to more positive potentials which improved the resolution of action potentials but decreased the signal size for synaptic potentials (Piao et al., 2015). The reduced optical signal response for sub-threshold potentials gives Bongwoori a better ''contrast'' for optically resolving action potentials.

A final design for researchers to consider when choosing a GEVI is the genetically encoded, hVOS (Chanda et al., 2005; Wang et al., 2010, 2012; Ghitani et al., 2015). First developed in the Bezanilla lab, hVOS consists of an FP anchored to the plasma membrane with the addition of a small charged molecule, DPA, that binds to the plasma membrane effectively acting as a fluorescent quencher. Since DPA is a lipophilic anion, the quenching agent will move from the outer surface of the plasma membrane to the inner surface upon membrane depolarizations Nakajima et al. Optogenetic Monitoring of Brain Activity

generating a voltage-responsive fluorescent signal. Like the other sensors, hVOS also has drawbacks which are primarily due to the fact that an exogenous chemical must be administered to the sample to be imaged. This is not a trivial process since too much DPA will significantly increase the capacitance of the plasma membrane and alter the neuronal activity of the cell. However, once the appropriate conditions are determined, hVOS gives optical signals for subthreshold potentials as well as action potentials in slice from populations of cells (Wang et al., 2012) or individual cells when expression of the FP is sparser (Ghitani et al., 2015).

### SYNAPTIC ACTIVITY MONITORING IN SLICES WITH GEVIs

Brain slices are invaluable for studying in detail the cellular, molecular, and circuitry activity of neuronal functions (Ting et al., 2014). GEVIs can expand this information since every pixel potentially becomes an electrode. There are not many examples of synaptic potential recordings from GEVIs in slice. Most examples are proof-of-principle type of recordings in the original publication of a new sensor to demonstrate its potential. The VSFP family of GEVIs are the most published recordings in brain slice (Akemann et al., 2013; Scott et al., 2014; Carandini et al., 2015; Empson et al., 2015; Mutoh et al., 2015). Here, we compare optical synaptic recordings in brain slices from VSFP Butterfly 1.2 and hVOS.

### FRET Signals of Butterfly in Cortical Brain Slices

**Figure 3A** shows the population imaging in coronal cortical slices prepared from a mouse brain electroporated in utero with VSFP-Butterfly 1.2 (Akemann et al., 2012). To explore voltage imaging from populations of cells, cortical slices were imaged at low magnification while delivering a single electrical stimulus (**Figure 3A**, left panel). The amplitude of the evoked optical signal ranged from 1 to 1.5% ∆R/R<sup>0</sup> (**Figure 3A**). Disinhibition with 25 mM gabazine increased the signal to 11% ∆R/R<sup>0</sup> (Akemann et al., 2012). Since VSFP-Butterfly 1.2 is a FRET probe, the ratio of the fluorescent change can be reported, but in slice the advantage of a ratiometric recording is of lesser value since movement artifacts due to respiration and blood flow do not exist. Despite this advantage, the voltage-dependent change in fluorescence is quite small, less than 0.5% ∆F/F which requires multiple trials to improve the signal to noise ratio.

### hVOS Signal in the Hippocampal Slice

**Figure 3B** shows the hVOS signal in a hippocampal slice. The electrical stimulation evoked clear fluorescence changes only when 4 µM DPA was present. This concentration of DPA provided excellent signal up to 2 h with minimum pharmacological action (Wang et al., 2010). The hVOS probe fluorescence decreases with membrane depolarization because DPA moves to the inner surface of the cell membrane where hVOS probes are anchored; the arrival of DPA quenches the probe fluorescence. Responses of approximately 1–3%

FIGURE 3 | Comparison of voltage indicators for synaptic imaging in brain slices. (A) Fluorescence and ratiometric signals of voltage-sensing fluorescent protein (VSFP) Butterfly 1.2 in cortical brain slices (modified with persmission from Akemann et al., 2012, Figures 3I,J). Left: wide-field fluorescence image with indicated position of the stimulation electrode (blue arrow). Scale bar, 150 µm. Middle and right: a single-pulse synaptic stimulus (middle: 100 µA; right: 200 µA) induced a depolarizing response as indicated by a transient decrease in mCitrine (yellow) and increase in mKate2 (red) emission from the pixels indicated by a red circle. The ratio of the two emission spectra is in black. (B) Stimulus-evoked fluorescence changes (∆F/F) and their dipicrylamine (DPA) dependence (modified with permission from Wang et al., 2012, Figure 3). A slice expressing hVOS 1.5 from mouse hippocampus. All recordings were from the striatum-radiatum (sr) of the CA1 region. Left: a slice from hVOS 1.5 line 602 with images and fluorescence traces superimposed. Traces are from the three numbered locations before and after addition of DPA. The stimulation site is indicated by the asterisk. All traces of hVOS signals are averages of 10 trials.

could be seen throughout the field of view (Wang et al., 2012).

VSFP-Butterfly 1.2, and hVOS can all generate an optical signal corresponding to synaptic responses in acute brain slices. hVOS has the larger ∆F/F. VSFP-Butterfly 1.2 does not require additional drug application to detect voltage changes in the neuron. Other sensors can also give optical signals in slice but those recordings have focused on action potentials in individual cells and are not shown here. The brightness, signal size, and voltage range of ASAP-1 make it a potentially useful sensor for imaging synaptic potentials in slice. While there are no reports in the literature of ArcLight being used to analyze neuronal activity in brain slices, the brightness, signal size and voltage-sensitivity are also ideal for optically recording synaptic potentials.

### SYNAPTIC ACTIVITY MONITORING IN VIVO WITH GEVIs

While slice recordings are extremely valuable for deciphering neuronal circuitry, the ultimate goal of voltage imaging is to detect neuronal activity in a behaving animal. This is an ambitious endeavor with very few examples, but some GEVIs

#### FIGURE 4 | Continued

images of ArcLight demonstrate membrane localization (arrow). The FP, mCherry, is localized to the nucleus to facilitate identification of transduced neurons. (d–f) Low magnification of the olfactory bulb—onl, olfactory nerve layer; gl, glomerular layer; epl, external plexiform layer; mcl, mitral cell layer. (g) Wide-field resting fluorescence intensity. (h) Glomerular patterns of activation after odor stimulation. (i) Odor-evoked optical signals from the region of interest marked with a red circle in (g,h). (j) Six unfiltered single trials aligned to the first sniff of odorant.

are now capable of giving a robust signal that allows in vivo imaging.

Proof of principle for in vivo voltage imaging was established by the Knöpfel lab using the VSFP family of probes (Akemann et al., 2012, 2013). **Figure 4A** shows single trial responses in the barrel cortex during whisker stimulation. Clearly, a stimulus evoked voltage signal could be detected in single trials even though the signal size is very small. Asterisks denote potential spontaneous voltage transients. However, unlike the stimulus evoked optical response, these potential transients exhibit different start times and kinetics. Having a low signal to noise ratio undermines the confidence in reliably detecting neuronal activity trial to trial (Carandini et al., 2015). Another drawback with VSFP Butterfly 1.2 is that the V1/<sup>2</sup> is roughly −70 mV with maximal fluorescent change occurring at −40 mV (**Figure 1**) making it virtually impossible to distinguish synaptic activity from action potentials based solely on signal size.

ArcLight has also been tested in vivo in mice and flies (Cao et al., 2013; Storace et al., 2015a). The V1/<sup>2</sup> for ArcLight is around −30 mV making it ideal to detect neuronal activity in flies whose action potentials range from a resting potential of −40 mV to a final excitation of −10 mV. As a side note this is why our probe, Bongwoori, should not be used for imaging neuronal activity in flies since the V1/<sup>2</sup> has been shifted to around 0 mV (Piao et al., 2015). **Figure 4B** shows a recording from a mouse expressing ArcLight in the olfactory bulb. As can be seen, respiration causes an optical artifact, but since ArcLight gives a large signal, identifying regions of the olfactory bulb responding to an odor is still possible. Again, though, it is not possible to resolve synaptic activity from action potentials.

The rhodopsin-based GEVIs have also been shown to elicit an optical signal in vivo. An example is shown in **Figure 5A** from the Gradinaru lab (Flytzanis et al., 2014). With the dim fluorescence of the GEVI, Archer, C. elegans is one of the few multicellular organisms one would be able to record from. After odorant stimulation, there is a slight variation in the ∆F compared to control. While there appears to be a slight signal, the low signal to noise ratio again undermines one's confidence in being able to reliably detect neuronal activity from trial to trial.

The last example of in vivo recordings is the best example of resolving action potentials. Ace2N-4AA-mNeon has been imaged in flies and mice (Gong et al., 2015). This probe is extremely fast showing the best fit of optical data to voltage yet (**Figure 5B**). The red arrow shows an optical response to a 5 mV

mouse expressing VSFP-Butterfly 1.2. White rectangle indicates area imaged in (c). Scale bar, 2 mm. (c,d) Ratio images obtained at times before and after brief deflection of the D1 whisker performed at time 0. (c) 10 ms before (d) 90 ms after the deflection. (e) Single-sweep (1, 2, and 3) mCitrine (yellow) and mKate2 (red) optical signals sampled from the region of interest indicated by white circle in (c), together with the 10-trial average (bottom). Asterisks mark signals corresponding to potential spontaneously occurring voltage transients. (f) Ratio (∆R/R0) signals corresponding to the traces in (e). (B) Odor-evoked signals of ArcLight in olfactory bulb (modified with permission from Storace et al., 2015a, Figures 1, 2). (a–c) High magnification confocal (Continued)

synaptic imaging in vivo. (A) Archer1 expressed in worms. (a) C. elegans expressing Archer1 shows fluorescence (λ = 655 nm; I = 880 mWmm−<sup>2</sup> , 100 ms exposure). Scale bar, 20 mm. (b) Top: experimental conditions: worms are stimulated with odorant (Isoamyl alcohol, IAA) for 5 min, flow is switched to buffer (S Basal) for 30 s, and then odorant flow is restored. The control conditions are performed on the same worm. Bottom traces: imaging of Archer1 fluorescence (250 Hz) (modified with permission from Flytzanis et al., 2014, Figures 5B,C). (B) Imaging single action potentials and subthreshold membrane voltage by Ace2N-4AA-mNeon (modified with permission from Gong et al., 2015, Figures 2A, 3D). (a) Optical resolution of action potentials of cultured hippocampal neurons under current clamp exhibiting best fit of optical data to electrical recording to date. Arrow denotes 5 mV depolarization. (b) Optical traces from a cortical V1→LM neuron in an awake mouse, showing visually evoked responses to drifting gratings.

depolarization. However, comparing the optical signal at the spike to the subthreshold potential, one can see that the optical response is skewed towards action potential activity. This gives a fantastic response when imaging the visual cortex in response to visual stimuli. Action potentials are easily discernible. The ability to optically report synaptic activity is less clear but still promising.

### CONCLUSION

GEVIs come in many flavors. As demonstrated, the signal size, speed, and voltage sensitivity affect the neuronal activity a GEVI can resolve. Many probes will give an optical signal in slice and in vivo but some signals will be more informative. If

the experimenter wants to image any neuronal activity from a population of cells in brain slice, the recommendations would be hVOS, ArcLight, and ASAP-1. All have a broad voltage range, traffic to the membrane well and give relatively large signals. ArcLight and ASAP-1 will have some difficulty in separating synaptic activity from action potentials due to their voltage sensitivities, but this could theoretically be overcome by coexpression of a red calcium sensor to verify action potential activity if the neuron tested has an action potential-induced calcium transient. If one wants to measure neuronal activity of individual cells in slice, then one should also consider Ace2N-4AA-mNeon.

Imaging single cells vs. a population of cells will also affect the choice of GEVI to be used. When imaging single cells, probes with broad voltage ranges will enable the optical detection of inhibition, synaptic potentials, and action potentials. However, these same probes when imaging large populations of cells are potentially less informative since the depolarization of a subgroup of neurons could swamp the small, hyperpolarizing signals from inhibited neurons.

Inefficient trafficking or high intracellular expression will affect the voltage imaging of a population of cells more so than when imaging individual cells. The reason for this is that the spatial representation of the cell under high magnification onto the pixels of the camera has changed. Under high magnification, a researcher can choose only pixels that correspond to regions of the cell that exhibited a fluorescent response. When imaging a population of cells, a pixel will be less likely to capture only the responsive fluorescence. This situation is depicted in **Figure 6**. When imaging a single cell, it is much easier to avoid the internal, non-responsive fluorescence and maximize the signal to noise ratio.

While the GEVIs currently available have shown significant improvement in their ability to optically detect neuronal activity, there is still much room for improvement. Refining the voltagesensitivity will enable maximizing the optical signal. For instance, a probe that only responded to hyperpolarization of the plasma membrane would make identifying the inhibited parts of a neuronal circuit much easier. Improving the membrane expression of the GEVI will decrease the nonresponsive fluorescence in a population of cells, thereby improving the signal to noise ratio. Most efforts to improve trafficking involve the addition of endoplasmic reticulum and Golgi release motifs. Codon optimization is another approach which for membrane proteins may be a misnomer. The idea of codon optimization

### REFERENCES


is to use only the most abundant codons for rapid translation of the protein. This has been shown to be effective for cytoplasmic proteins. However, for membrane proteins slowing the translation to allow proper folding and insertion into the translocon may also be important (Norholm et al., 2012; Yu et al., 2015). Finally, limiting the expression of the GEVI to subcellular components (i.e., the soma, dendrites, etc.) could also focus the optical signal to the desired region of the neuron again improving the signal to noise ratio.

### AUTHOR CONTRIBUTIONS

RN and AJ wrote the manuscript and contributed figures. B-JY and BJB helped to write the manuscript.

### ACKNOWLEDGMENTS

This work was funded by the Korea Institute of Science and Technology (KIST) Institutional Program Multiscale Functional Connectomics, 2E26190 and KIST Institutional Program 2E26170.


neurons with a fluorescent protein voltage probe. Neuron 75, 779–785. doi: 10. 1016/j.neuron.2012.06.040


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Nakajima, Jung, Yoon and Baker. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

## Control of Transmembrane Protein Diffusion within the Postsynaptic Density Assessed by Simultaneous Single-Molecule Tracking and Localization Microscopy

### Tuo P. Li and Thomas A. Blanpied\*

Department of Physiology and Program in Neuroscience, University of Maryland School of Medicine, Baltimore, MD, USA

Postsynaptic transmembrane proteins are critical elements of synapses, mediating trans-cellular contact, sensitivity to neurotransmitters and other signaling molecules, and flux of Ca and other ions. Positioning and mobility of each member of this large class of proteins is critical to their individual function at the synapse. One critical example is that the position of glutamate receptors within the postsynaptic density (PSD) strongly modulates their function by aligning or misaligning them with sites of presynaptic vesicle fusion. In addition, the regulated ability of receptors to move in or out of the synapse is critical for activity-dependent plasticity. However, factors that control receptor mobility within the boundaries of the synapse are not well understood. Notably, PSD scaffold molecules accumulate in domains much smaller than the synapse. Within these nanodomains, the density of proteins is considerably higher than that of the synapse as a whole, so high that steric hindrance is expected to reduce receptor mobility substantially. However, while numerical modeling has demonstrated several features of how the varying protein density across the face of a single PSD may modulate receptor motion, there is little experimental information about the extent of this influence. To address this critical aspect of synaptic organizational dynamics, we performed single-molecule tracking of transmembrane proteins using universal point accumulation-for-imaging-in-nanoscale-topography (uPAINT) over PSDs whose internal structure was simultaneously resolved using photoactivated localization microscopy (PALM). The results provide important experimental confirmation that PSD scaffold protein density strongly influences the mobility of transmembrane proteins. A protein with a cytosolic domain that does not bind PSD-95 was still slowed in regions of high PSD-95 density, suggesting that crowding by scaffold molecules and perhaps other proteins is sufficient to stabilize receptors even in the absence of binding. Because numerous proteins thought to be involved in establishing PSD structure are linked to disorders including autism and depression, this motivates further exploration of how PSD nanostructure is created. The combined application PALM and uPAINT should be invaluable for distinguishing the interactions of mobile proteins with their nanoenvironment both in synapses and other cellular compartments.

#### Edited by:

George Augustine, Nanyang Technological University, Singapore

Reviewed by:

Michele H. Jacob, Tufts University, USA Jason D. Shepherd, University of Utah, USA

\*Correspondence: Thomas A. Blanpied tblanpied@som.umaryland.edu

> Received: 30 April 2016 Accepted: 05 July 2016 Published: 22 July 2016

#### Citation:

Li TP and Blanpied TA (2016) Control of Transmembrane Protein Diffusion within the Postsynaptic Density Assessed by Simultaneous Single-Molecule Tracking and Localization Microscopy. Front. Synaptic Neurosci. 8:19. doi: 10.3389/fnsyn.2016.00019

Keywords: uPAINT, PALM, synapse, nanodomain, glutamate receptor, macromolecular crowding

## INTRODUCTION

Transmembrane proteins such as receptors diffuse on the cell surface to reach their sites of action. In doing so, they must make their way through complex environments typified by varying densities of obstacles and potential binding partners. The average behavior of proteins moving through such environments has been well characterized (Frick et al., 2007). However, on small spatial scales or within small compartments, the local organization of potential interactors will dominate the influence on receptor motion paths (Kusumi et al., 2014). For instance, locally high concentrations of steric obstacles create a phenomenon called macromolecular crowding (Ryan et al., 1988) that can slow mobility and result in anomalous diffusion (Saxton, 1994; Santamaria et al., 2010). Thus, high-resolution information about the distribution of even non-binding obstacles is necessary to understand motion trajectories of transmembrane proteins on small scales.

Perhaps the most complex compartment of the plasma membrane in neurons is the postsynaptic density (PSD). The PSD of glutamatergic synapses concentrates numerous receptor types aligned to the presynaptic active zone. Despite the small size of the average PSD (∼0.08 µm<sup>2</sup> × 50 nm; Harris and Weinberg, 2012), roughly 500 species of proteins can be found in this compartment (Husi et al., 2000; Sheng and Hoogenraad, 2007). Because of the high local density of transmembrane proteins, receptor-binding proteins such as PSD-95, and juxtamembrane cytosolic molecules, protein motion within the membrane at the synapse is likely extremely obstructed (Santamaria et al., 2010). This complicated environment is critical to understand, because protein organization in the PSD directly regulates synaptic transmission in many ways. The number of glutamate receptors present in the PSD sets an upper limit on the strength of the synapse (Huganir and Nicoll, 2013), and receptors exchange continuously by diffusion between the PSD and the perisynaptic plasma membrane (Opazo and Choquet, 2011; Choquet and Triller, 2013). Further, alterations to the PSD are a critical component of activity-driven plasticity mechanisms regulating receptor number (Inoue and Okabe, 2003; Bosch et al., 2014). Thus, understanding mechanisms within the PSD that control motion of glutamate receptors is critical for determining how receptor number is modulated during plasticity.

Even beyond the clear importance of the number of receptors, however, their distribution within the synapse in the plane of the membrane is a vital regulator of synaptic strength (MacGillavry et al., 2011). This is because when glutamatergic vesicles fuse with the presynaptic plasma membrane, the result is a highly concentrated but narrow spike of released neurotransmitter. The rapid dissipation of this spike by glutamate diffusion means that receptors laterally displaced from the site of fusion even by less than 100 nm often fail to activate (Xie et al., 1997; Raghavachari and Lisman, 2004; Santucci and Raghavachari, 2008; Freche et al., 2011). Amplifying this effect, receptors in the PSD are concentrated in ∼80 nm subdomains (MacGillavry et al., 2013; Nair et al., 2013) where the principle receptor-binding scaffold PSD-95 is also concentrated (Fukata et al., 2013; MacGillavry et al., 2013). In previous work, we modeled diffusion within PSDs where the heterogeneous distribution of PSD-95 was measured, and found that the clustered nature of this scaffold could strongly limit the ability of transmembrane proteins to enter (or escape) the crowded regions of the PSD (Li et al., 2016). Thus, nanoscale regional variation in protein composition within a single PSD may have strong impact on synaptic transmission by controlling the subsynaptic distribution of receptors.

A major impediment to progress on this issue is the technical challenge of simultaneously measuring the nanoscale distribution of the protein environment while simultaneously measuring protein motion through it. To address this, we developed a combined single-molecule imaging approach that uses single-particle tracking photoactivated localization microscopy (sptPALM; Manley et al., 2008) to map the positions of PSD-95 molecules within the synapse (MacGillavry et al., 2013), while simultaneously tracking the motion of proteins in the plasma membrane by universal point-accumulation-forimaging-in-nanoscale-topography (uPAINT; Giannone et al., 2010). Using this strategy, we could directly investigate the influence of both obstacle density and protein binding on motion through the PSD. The results provide direct experimental confirmation that macromolecular crowding within the PSD can strongly limit the motion of even small transmembrane proteins, likely helping to establish the distribution and dynamic exchange characteristics of glutamate receptors and other molecules.

### MATERIALS AND METHODS

### Neuron Culture and Transfection

Dissociated hippocampal neuron cultures were prepared from E18 rat embryos as described previously (Frost et al., 2010). All procedures conformed to the guidelines established by Animal Welfare Act, Public Health Service, and the United States Department of Agriculture, and were approved by the Institutional Animal Care and Use Committee at the University of Maryland, Baltimore. Prior to plating the cells on coverslips, the coverslips were first cleaned as reported previously (MacGillavry et al., 2013), subsequently coated with lateral-drift tracking, yellow-green fluorescent 100 nm beads (F8803; Thermo Fischer Scientific) diluted 1:25,000 in 100% ethanol (dried within 25 min in the hood), and then coated overnight with poly-L-lysine (Sigma). Cells were transfected at DIV10–13 using Lipofectamine 2000 (Thermo Fischer Scientific) and imaged 72–96 h later (unless stated otherwise). Individual coverslips were transfected with 0.5–0.75 µg of cDNA for each expression construct.

### Expression Constructs

Plasmid cDNAs were obtained as follows (with original sources): the binding and nonbinding probes, Super ecliptic phluorin (SEP)-transmembrane (TM)-Bind and SEP-TM-Nonbind (Li et al., 2016), the PSD-95-mEos2 replacement plasmid shrPSD-95-mEos2 (MacGillavry et al., 2013). We have previously measured that this construct produces a mild 1.4× overexpression. Interpretations of the results from the PALM-PAINT assay require control experiments to ensure that the expressed protein is organized similarly to its endogenous counterpart. To this end, we have also compared the PSD-95 nanostructure of untransfected cells and those expressing the replacement construct PSD-95-mEos2 (MacGillavry et al., 2013; Tang et al., in press). These were not different in terms of PSD area, number of subsynaptic nanoclusters, and the size of those nanoclusters. Thus, we believe the mEos2 tag does not affect PSD nanostructure, at least in the absence of severe PSD-95 overexpression. Furthermore, mEos2 is fused at the c-terminal end of the PSD-95, not altering the positions of the various proteinprotein interaction motifs. Thus, we believe tagging itself does not affect the interaction of PSD-95 with other proteins.

### Two-Color Single-Molecule Imaging

PALM-PAINT, a combination of PALM (Betzig et al., 2006; Hess et al., 2006) and uPAINT (Giannone et al., 2010), was performed through a Photometrics DV2 on an Olympus IX81 ZDC2 inverted microscope that was described by MacGillavry et al. (2013). Cells expressing the indicated constructs were imaged in a previously described extracellular buffer (Li et al., 2016). SEP-containing probes were labeled with ATTO647Nconjugated anti-green fluorescent protein (GFP) nanobodies (GFPBooster-647N, Chromotek), bath applied to a final concentration of 0.5–2 nM once the first stretch of synapses was identified for each coverslip. Cells remained at 25◦C for no more than 30 min per imaging session.

We imaged the red and far-red bands by interleaving excitations of 561 and 640 nm. Imaging was conducted at 29 Hz (14.5 Hz per color), with 10-ms duration excitation per frame for 5000–20,000 frames per color. The two emissions were overlaid based on calibration images of TetraSpeck beads (100 nm; Thermo Fischer Scientific) deposited on an acellular coverslip as described by MacGillavry et al. (2013). Yellow-green beads (100 nm) ethanol-diluted and dried onto the coverslips prior to plating the cells (F8803; Thermo Fischer Scientific) were excited and captured once every 1000 frames to monitor lateral drift. To correct lateral drift, we localized fiducials post hoc from images of the yellow-green beads. We screened for spurious localizations by the duration of fluorescence and mobility. Namely, the bead ought to be present on the first frame, persist for as long as each imaging session, and displace <100 nm (1 pixel) per 1000 frames. Such a filtering process provided a list of localizations that correspond to fiducials. From this list we calculated the sample lateral drift as the weighted average of the displacements of all fiduciary localizations between each set of 1000 frames. For weights, we used the inverse of the estimated localization uncertainty (Thompson et al., 2002) of each fiducial. The single linear correction in drift we applied to each subset of 1000 frames was the average correction obtained from the estimates of 2–10 fiducials in the field of view.

## Single-Molecule Localization, Tracking Analysis, and PSD Nanostructure Analysis

All data analysis was performed offline using custom routines in MATLAB (The MathWorks). The algorithms for determining molecule location and criteria for filtering molecules to be considered for further analysis were applied as previously described (MacGillavry et al., 2013). In addition to filtering by localization precision, elliptical form, and brightness, we also utilized a Voronoi-based segmentation program SR-Tesseler (Levet et al., 2015) to filter out spurious localizations outside of putative neuronal border. Criteria for defining a track were described by Li et al. (2016). Instantaneous effective diffusion coefficients (Deff) at individual track time points were calculated for tracks that persisted at least eight frames (the duration in which the mean mean squared displacement (MSD) was linear; a more detailed description can be found in Lu et al., 2014). For comparing diffusion inside and outside of PSDs, tracks that entered or exited the PSD were divided into two portions, a synaptic and an extrasynaptic subtrack, the Deff of which were calculated by averaging the instantaneous Deff for the tracked localizations therein. For the rare tracks that entered and exited PSDs multiple times, synaptic Deff was determined by averaging the instantaneous Deff of all the tracked locations inside the PSD border; vice versa to calculate the extrasynaptic Deff. To calculate the Deff of subtracks in other cases (e.g., within a particular range of PSD-95 regional density, or below the detection limit of Deff), we calculated the average of instantaneous Deff of localizations meeting the criteria. The lower detection limit of Deff was determined conservatively by calculating the Deff of a theoretical immobile particle displaced as much as the average trajectory error ∼30 nm (Savin and Doyle, 2005; Lu et al., 2014) per time frame, which amounted to 0.003 µm<sup>2</sup> /s; this value is indicated by a gray vertical dotted or dotted-dashed line in cumulative frequency graphs of Deff.

The PALM PSD border was determined by taking the convex hull of the PSD-95 molecular positions. To determine the Gaussian-blurred border of each PSD, we first constructed a 2-dimensional molecular density map of 25 × 25 nm subpixels from PSD-95 molecules for each PSD. We then convolved it with a constant-amplitude Gaussian image profile (σ = 125 nm, image width = 900 nm) that is similar to an ideal microscope point-spread function with a full-width at half max (FWHM) of ∼250 nm. To determine the FWHM border of the blurred PSD, we thresholded the convolved image at half-maximum intensity. To determine the 95%-border, we thresholded it at 5% of the maximum intensity.

### Determination of Regional PSD-95 Density

To quantify the regional density of PSD-95, we measured the number of molecules surrounding each position of a tracked probe molecule. PSD-95 localizations appearing in consecutive frames separated by no more than 200 nm were considered one molecule, and its position was taken from the first frame it appeared. We counted the number of PSD-95 molecules within a 30 nm radius (the average trajectory error) of each position in the track of a probe molecule. Because we wished primarily to relate Deff to regional density, we used the same tracked locations to calculate each measure. That is, for the regional density of PSD-95 surrounding the probe at each position, we calculated the forward running average of regional densities for eight frames, the same number of frames used to calculate Deff. The absolute density of molecules in all calculations was adjusted by the average expected number of blinks (one) of mEos2 in our experimental conditions (Annibale et al., 2011; MacGillavry et al., 2013). The final density when averaged across synapses is likely somewhat higher than endogenous PSD-95 densities due to the mild (∼1.4-fold) overexpression obtained with our shRNA knockdown/replacement approach. However, we did not attempt to measure the potentially quite variable ratio of endogenous and tagged PSD-95 at each synapse analyzed, and did not take this into account for calculations. Data of immunostaining done in cells not intended for PALM-PAINT showed that endogenous PSD-95 molecules resistant to the knockdown did not form dense clusters outside those formed by the expressed PSD-95 molecules, indicating that though the endogenous PSD-95 might be present in the synapse, they likely raise the absolute number of PSD-95 molecules without changing the spatial variation in regional density.

### Statistics

Where means are presented, the accompanying errors are the standard error of the mean; additionally, these data were normally distributed according to the Shapiro-Wilk normality test. Where box-and-whisker plots are presented, the middle bar represents the median, the upper and lower limit of the boxes denote the interquartile range, and the whiskers extend to 5% and 95% of the distribution; additionally, these data were not normally distributed according to the Shapiro-Wilk normality test. Different sets of statistical tests were used for normally and non-normally distributed data. Pairwise statistical tests were performed using unpaired t-test with Welch's correction for normally distributed data; they were performed using Mann-Whitney U test for non-normally distributed data. Where two-way analysis of variance (ANOVA) was used, a Bonferroni correction was used for post hoc pairwise comparisons. Kolmogorov-Smirnov tests were applied for cumulative frequency distributions. In all cases, means (or medians) were considered significantly different if the test reported p < 0.05. Most statistical tests and all graphing were done using Prism (GraphPad Software). Two-way ANOVA was done in MATLAB (The MathWorks).

### RESULTS

To perform single-molecule tracking during superresolution imaging of the PSD, we co-transfected 13–17 DIV hippocampal neurons with two cDNA constructs. The first encoded a single-pass TM protein composed of an extracellular SEP, the TM domain, and the intracellular carboxy terminus of stargazin, which enables this protein to bind PSD-95. Thus, we refer to it as SEP-TM-Bind (Li et al., 2016). This was co-transfected with shrPSD-95-mEos2, which expresses shRNA

FIGURE 1 | Photoactivated localization microscopy (PALM)-point accumulation-for-imaging-in-nanoscale-topography (PAINT), single-molecule tracking during PALM imaging. (A) (Left) Super ecliptic phluorin (SEP)-TM-Bind molecules tracked for at least eight frames super-imposed on molecules of shrPSD-95, accumulated from 5000–20,000 frames. (Right) A typical example of tracked probes superimposed on positions of shrPSD-95 molecules. The first and last localized positions are indicated as filled and open circles, respectively. (B) Mean squared displacement (MSD) over time of connected sub-segments of tracks (subtracks) that are lasted at least 15 frames (n = 156 synaptic and 2907 extrasynaptic tracks/113 PSDs/13 fields/11 cells/3 cultures). (C) Cumulative frequency distributions of Deff for subtracks that are at least eight frames long (n = 654 synaptic and 3349 extrasynaptic tracks/113/13/11/3). ∗∗∗∗p < 0.0001.

targeting PSD-95 along with an RNAi-resistant, mEos2-tagged PSD-95 (MacGillavry et al., 2013). To track motion of SEP-TM-Bind at the cell surface, we used uPAINT and applied anti-GFP nanobodies carrying Atto647N in the chamber to a final concentration of 0.5–2 nM. Concurrently, we localized the positions of PSD-95-mEos2 using conventional sptPALM methods (**Figure 1A**).

The SEP-TM-Bind probe exhibited clearly different mobility inside and outside of the PSD (**Figure 1A** right), as expected based on the behavior of AMPA-type glutamate receptors (Bats et al., 2007; Hoze et al., 2012), intercellular adhesion molecules such as neuroligin and leucine rich repeat transmembrane neuronal 2 (LRRTM2; Chamma et al., 2016), and NMDA-type glutamate receptors (Dupuis et al., 2014). We compared the diffusion patterns of molecules inside and outside of the PSD, when the PSD border was defined by the convex-hull border of PSD-95 positions. Probes outside of the PSD displayed near-free diffusion as evidenced by an almost linear plot MSD (**Figure 1B**). However, probes within the PSD showed a much slower mobility and appeared highly confined in their motion, as evidenced by saturation of the relationship between MSD and time. This curve approached a plateau of 104 ± 35 nm<sup>2</sup> , suggesting confinement within <60 nm diameter regions of the PSD (Ehlers et al., 2007). The effective diffusion coefficients (Deff) of the synaptic subtracks were ∼2 orders of magnitude slower than those of the extrasynaptic subtracks (**Figure 1C**). This differential is similar to that seen for AMPARs (Bats et al., 2007; Hoze et al., 2012), suggesting that AMPAR motion within synapses likely is regulated by mechanisms that also impact many other types of molecules.

The improved resolution of the PSD border obtained by imaging the positions of individual PSD-95 molecules as opposed to using widefield or confocal microscopy should improve discrimination of which molecules are within the synapse. The average PSD area (0.085 ± 0.006 µm<sup>2</sup> , n = 263 PSDs/21 neurons) was within the ranges as previously detected by PALM (MacGillavry et al., 2013; Nair et al., 2013) and by electron microscopy (Harris and Stevens, 1989; Schikorski and Stevens, 1997; Shinohara et al., 2008). To test whether diffraction-limited PSD borders could perform as well as PALM of PSDs at segregating synaptic and extrasynaptic probes, we simulated diffraction by blurring the PSD-95 molecular density maps with a Gaussian point-spread function. Taking the FWHM of this intensity distribution as the border of the diffractionblurred PSDs, as was done by Li et al. (2016), performed nearly as well as the PALM'ed PSD border in segregating synaptic and extrasynaptic probe movements. Taking the 95% border of the blurred PSDs diminished the difference between synaptic and extrasynaptic Deff (**Figure 2**). Thus, how the PSD border is defined in diffraction-limited approaches can influence the accuracy of segregating diffusing molecules in different sub-compartments of the cell.

### Synaptic TM Protein Diffusion is not Influenced by PSD Size or Whole-Synapse PSD-95 Density

Interestingly, the Deff distribution within different synapses varied widely, and individual molecules exhibited Deff spanning more than five orders of magnitude. This difference did not stem from neuron-to-neuron variability, as the median synaptic Deff of the binding probes measured in different neurons differed by only up to 2-fold. We reasoned that this broad range of Deff may arise because the diffusion environment within the PSD might vary based on synaptic size or geometry, or because the density of binding sites could influence how likely a probe is able to be bound at any given time. To test this, we first examined the relationship between the area of the PSD and the median Deff of probes found within it. Based on linear regression analysis, we found no statistically significant correlation (**Figure 3A**). However, our previous study found that the fluorescence recovery of these probes after photobleaching spines was negatively correlated with PSD area (Li et al., 2016). Combined with this finding, this suggests that the size of the synapse correlates with the rate with which these probes enter and exit the spine, but does not influence their diffusion within the synapse. To determine whether overall PSD-95 density within the synapse can determine the diffusion

of the binding probes, we examined the relationship between the density of PSD-95-mEos2 localizations and the synaptic median Deff of probes in each PSD. The absolute density of localizations in all calculations was adjusted by the average expected number of blinks (one) of mEos2 in our experimental conditions (Annibale et al., 2011; MacGillavry et al., 2013). It should be noted that this measure of density does not incorporate the unknown fraction of total PSD-95 molecules that were mapped, and also ignores the numerous other binding partners of SEP-TM-Bind that may not correlate with the

measured density of PSD-95-mEos2 as well as the likelihood of slight overexpression compared to endogenous protein level (∼1.4×, see MacGillavry et al., 2013). Nevertheless, we found no statistically significant correlation (**Figure 3B**), suggesting that, given these caveats, the overall density of PSD-95 within the PSD does not influence the median diffusion of binding TM proteins in the synapse.

### The Control of TM Protein Diffusion by Binding and Steric Hindrance Within the PSD

Though the overall measured density of PSD-95 did not correlate with the diffusion coefficient of SEP-TM-Bind, it would be surprising if this key scaffolding protein did not affect the mobility of its binding partners at all. We thus considered that the distribution of PSD-95 molecules is highly heterogeneous within single synapses (Fukata et al., 2013; MacGillavry et al., 2013; Nair et al., 2013; Broadhead et al., 2016) and can display multiple regions of high density within the synapse. Namely, two synapses of the same PSD-95 density can have very different arrangement of PSD-95 molecules, an organization that could obscure the effect that PSD-95 molecular density can have on the diffusion of probes when measured at the level of the entire synapse. Consistent with this notion, computer modeling has demonstrated that measured arrangements of PSD-95 molecules can prevent a larger fraction of TM proteins from escaping the synapse than homogeneously distributed PSD-95 molecules, without changing the overall density of PSD-95 (Li et al., 2016).

To test whether the density of PSD-95 immediately surrounding the probe can influence its diffusion within the synapse, we defined a subsynaptic metric termed ''regional PSD-95 density'' to be the number of PSD-95 molecules surrounding a tracked probe position. We measured the regional density using a fixed radius based on the average positional error of the tracked molecules (30 nm, see ''Materials and Methods'' Section). However, the results of the following analyses depended only very weakly on the radius over the range of 15–80 nm (data not shown). We subdivided tracks into subsegments (subtracks) based on the regional PSD-95 density at each of their positions, and plotted the Deff for subtracks based on their regional density. This analysis revealed that the higher the regional PSD-95 density, the slower the diffusion coefficients of the subtracks in that area of the PSD (**Figures 4A,B**). In fact, the median probe Deff within the synapse was strongly correlated with the regional density of PSD-95 (**Figure 4C**).

### The Control of TM Protein Diffusion by Steric Hindrance Alone Within the PSD

At a first glance, this result may not be surprising, as it supports the idea that the more scaffold binding partners there are in the synapse, the more likely the probe will be bound and thus immobilized before diffusing further. However, this effect is more difficult to interpret if we consider that PSD-95 not only binds this probe, but can serve as a steric obstacle. In fact, PSD-95 is a hub for binding many other proteins that can serve as additional obstacles which could, without binding the probe, hinder its diffusion. To isolate the effect of steric hindrance from the combined effect of steric hindrance and

probe Deff outside of PSDs (n = 3349 tracks for SEP-TM-Bind, 4470 SEP-TM-Nonbind). (D) (Left) PSD area and median Deff within each of the PSDs; linear regression test (n = 91 PSDs/10 cells/3 cultures of SEP-TM-Nonbind, same as in Figure 3 for SEP-TM-Bind). (Right) Overall synaptic PSD-95 density and median Deff within each of the PSDs; linear regression test (n = as in left panel).

probe-scaffold binding, we performed PALM-PAINT on a probe variant that cannot bind to PSD-95 (SEP-TM-Nonbind from Li et al., 2016). Interestingly, SEP-TM-Nonbind still entered synapses and diffused within them, but did not enrich within the PSD nearly as greatly as SEP-TM-Bind (**Figure 5A** and see also Li et al., 2016). Notably, the diffusion of this nonbinding probe within the synapse was dramatically faster than that of the binding probe (**Figure 5B**). On the other hand, the extrasynaptic diffusion of the nonbinding probe was not different from that of the binding probe (**Figure 5C**). It is interesting to note that using sptPALM (i.e., by photoactivating and tracking an mEos3 fusion protein rather than using an anti-GFP PAINT approach as here) there was a very similar mobility differential between the binding and nonbinding probes despite a substantial absolute difference in Deff arising from the poorer localization precision of the fusion protein compared to the organic dye (Li et al., 2016). Thus, the mobility difference between the two probes is quite robust. Intriguingly, the shoulder-like shape of the cumulative Deff distribution suggests that probes undergo multiple influences on their diffusion within the synapse. However, neither PSD area nor whole-synapse PSD-95 density correlated with probe diffusion within the PSD (**Figure 5D**).

By labeling SEP-tagged TM proteins with nanobodies, we add minimal but appreciable bulk to the extracellular domain of the diffusing entity. SEP is fused to the extracellular domain of the TM probes, adding approximately 3–5 nm of bulk; nanobody labeling of the GFP adds an additional 3–5 nm of bulk. While additional extracellular bulk has previously been shown to slow receptor diffusion (Groc et al., 2007), we expect the effect to be identical for both the binding and the nonbinding probes and thus not change our conclusions about the effect of postsynaptic steric hindrance.

We next considered whether the regional density of PSD-95 immediately surrounding the nonbinding probe can sterically control the probe diffusion. To test this, we first subdivided the tracks into subtracks and binned them into increasing regional densities of PSD-95 molecules, as in **Figure 4B**. This revealed that despite the lack of a PSD-95-binding motif, the probe still diffused more slowly within higher density regions of the PSD (**Figure 6A**). The effect appeared to saturate at low Deff since the mobility of these slowly moving molecules is below our detection limit (0.003 µm<sup>2</sup> /s).

If steric obstruction influences probe mobility, it may influence the overall pattern of probe position within the synapse. We thus compared the fraction of subtracks found in different regional PSD-95 densities (**Figure 6B**). Interestingly, though the binding and the nonbinding probes were distributed similarly through most density values, the nonbinding probes appeared to be preferentially excluded from the highly dense subregions (i.e., >25 regional molecules) of the synapse. Because of the small number of subtracks in these bins, however, this difference was not significant. Note also that the distribution of tracked molecules may not completely faithfully represent the total steady-state distribution of probe molecules, since molecules immobilized in the synapse for long periods are less likely to be recognized by a nanobody and be tracked by uPAINT.

It appeared that regional densities of PSD-95 higher than 10 had minimal effect on diffusion of the nonbinding probe in the Deff range below our detection limit (**Figure 6C**), whereas the regional density of PSD-95 linearly correlated with

FIGURE 6 | Subsynaptic regional density of PSD-95 influences the mobility of a probe that does not bind PSD-95. (A) Cumulative frequency distributions of the nonbinding probe Deff binned in increasing regional densites of PSD-95 surrounding the subtracks (n = 480 subtracks in 0–5 PSD-95, 204 in 5–10, 148 in 10–15, 81 in 15–20, 32 in 20–25, 12 in 25–30, 13 in 30+). (B) Fraction of subtracks in different regional densities of PSD-95. (C) PSD-95 regional density and the median Deff of the nonbinding probe. (D) Fraction of tracks with subsegments that were slowed below the detection limit per PSD (n = 113 PSDs for SEP-TM-Bind, 91 SEP-TM-Nonbind; <sup>∗</sup>p = 0.0123 Mann-Whitney U test). (E) (Left) Cartoon highlighting the PSD-95 molecules surrounding subtrack durations that diffused below the detection limit. The open and closed purple circles indicate the beginning and end of a track, the circles pseudo-colored by regional density were within 30 nm of sub-detection limit subtracks. We calculated the median regional density of PSD-95 of all subtrack durations that diffused below the detection limit with every PSD. (Right) Median regional density of PSD-95 surrounding subtracks that were below the detection limit (n same as in D; <sup>∗</sup>p = 0.0325 K-S test). (F) Median regional density of PSD-95 surrounding subtracks that above the detection limit (n same as in D; ns, Not significant, p = 0.158 K-S test).

the diffusion of the binding probe (**Figures 4B,C**). Thus we wondered whether the two probes required different degrees of steric hindrance in order to be stabilized. To test this, we first noted that during multiple visits to a synapse, or even within a single visit, probe molecules could display both slow and fast periods of motion. Considering only the subtracks within synapses, SEP-TM-Bind had a higher fraction than SEP-TM-Nonbind of these subtracks for which Deff was below the detection limit (**Figure 6D**). This indicates that the binding probes were more often slowed down while within the synapse than were the non-binding probes. We could then further exploit the ability afforded by PALM with uPAINT to examine the very local environment of the molecules as they moved within the synapse. Specifically, we noted that if steric hindrance slows mobility of the both the binding and non-binding probes, but binding is only able to slow SEP-TM-Bind, then SEP-TM-Bind would be expected to show a greater tendency to slow its mobility in relatively less dense PSD subregions. That is, even sparse binding partners could potentially capture and immobilize SEP-TM-Bind whereas higher concentrations of molecules would be required to sterically obstruct SEP-TM-Nonbind. Consistent with this, when we analyzed the sub-detection-limit portion of the nonbinding probe subtracks, we found that these were preferentially in locales of higher regional density of PSD-95 compared to those of the binding probes (**Figure 6E**). Interestingly, the fraction of the subtracks above the detection limit did not show any difference in regional PSD-95 density (**Figure 6F**), with even a trend to the opposite relationship.

Altogether, these results suggest that crowding by scaffold molecules and perhaps other proteins is sufficient to stabilize TM proteins in the absence of binding. How dense does the molecular environment have to be in order to slow the TM probes sterically as much as the combined influence of steric hindrance and probe-scaffold binding? To estimate an answer to this question, we compared the diffusion coefficients of the binding and nonbinding probes within increasing regional densities of PSD-95 (**Figure 7**). As expected based on **Figures 4**, **6**, the synaptic Deff of both probes decreased gradually with increasing PSD-95 regional density. However, the Deff of SEP-TM-Nonbind decreased precipitously over the range of 0–15; yet, it did not decrease further at higher densities. Furthermore, the Deff of SEP-TM-Nonbind plateaued at the Deff value displayed by SEP-TM-Bind at very low PSD-95 densities. Thus, by this analysis, ∼15 PSD-95 molecules per region of 30 nm radius (∼5000 molecules/µm<sup>2</sup> ) is the threshold beyond which the steric hindrance is as strong as both steric and binding influences combined.

### DISCUSSION

Using simultaneous single-molecule tracking and localization microscopy enabled by uPAINT and PALM, we demonstrated that the subsynaptic regional density of a scaffold protein PSD-95 can stabilize the surface membrane diffusion and positional organization of a single-pass transmembrane protein probe. The denser the regional density of PSD-95, the slower was

the diffusion of the TM probe. This influence was apparent even in the absence of probe-scaffold binding, indicating steric hindrance by macromolecular crowding can complement protein-protein binding interactions in organizing TM proteins within the synapse.

### The Roles of Receptor-Scaffold Binding and Macromolecular Crowding in Subsynaptic Organization

The mobility of AMPARs in the synapse is increased when the binding of their accessory subunit Stargazin to PSD-95 is disrupted (Bats et al., 2007; Sainlos et al., 2011), providing strong evidence that receptors are acutely stabilized by PSD-95 binding. However, even in these conditions some stabilization of receptors in the synapse occurs, and our results indicate specifically that even small probes carrying a cytosolic tail unable to bind PSD-95 is still slowed substantially in the synapse. What controls this stabilized fraction even in the absence of binding has been a mystery. Mechanisms such as additional binding interactions have been proposed, which is not unlikely considering that AMPARs have numerous auxiliary subunits (Tomita et al., 2003; Cho et al., 2007; Soto et al., 2009; Kalashnikova et al., 2010; von Engelhardt et al., 2010; Erlenhardt et al., 2016) that can bind to various scaffolding proteins. However, even a cytosolic domain composed of just a GFP-type molecule is slowed within the synapse (Li et al., 2016). Thus, we propose a more general mechanism that likely applies not only to glutamate receptors but also to other TM proteins critically important for synaptic function. In this model, receptor-scaffold binding is a ticket to entry and exit; PSD morphing redistributes even bound receptors within the synapse; and macromolecular crowding in combination with binding stabilizes the receptors at subsynaptic domains highly packed with other proteins important for synaptic transmission.

Interestingly, though the TM binding probe is much smaller than AMPARs, its distribution of Deff within the synapse appeared as slow as, or even slower than, AMPAR diffusion measured in other studies using dye-conjugated antibodies or quantum-dot-conjugated nanobodies (Nair et al., 2013; Cai et al., 2014). It is possible that these various labeling approaches preferentially sample receptors that exit the synapse and diffuse more freely in the perisynaptic space, which may tend to obscure real differences in the relative numbers of low-Deff molecules that occupy the synapse for long periods. On the other hand, taken at face value, the similarity is consistent with our previous report using sptPALM that a substantial and nearly identical fraction of mEos-tagged AMPARs and binding probes diffused with Deff < 0.02 µm<sup>2</sup> /s (Li et al., 2016), further supporting the notion that much of the synapse is so crowded it stabilizes and organizes both large and small TM proteins.

Previously, we demonstrated using partial synapse Fluorescence Recovery after Photobleaching (FRAP), smtPALM alone, and uPAINT alone that the protein bulk of the TM probe decreased its diffusion within the synapse. However, previous approaches were limited in ability to assess how the complex structure within the PSD could influence the mobility and organization of TM proteins in a living synapse. Results from PALM-PAINT indicate that a particular degree of postsynaptic crowding, which we estimate as 5000 molecules/µm<sup>2</sup> , can be sufficient to stabilize TM protein diffusion in the absence of binding. This density translates to an average ∼14 nm inter-PSD-95 distance, very similar to the mean nearest-neighbor distance of ∼13 nm between the ''vertical filaments'' corresponding to PSD-95 as measured in EM tomography (Chen et al., 2008). Note that we deduce this as an average spacing, but could not directly measure it around individual moving probes. The similarity between these values suggests that rather subtle variations in scaffold density across the lateral extent of the synapse could change TM protein mobility substantially. This high density packing is similar to what has been measured for AMPA receptors (e.g., 2000–4000/µm<sup>2</sup> , see Levet et al., 2015). Indeed, receptor-scaffold binding may facilitate the assembly of this tight packing. Though the fractional time synaptic AMPARs spend bound to PSD-95 is not known, macromolecular crowding is likely to augment maintenance of this architecture once assembled, because receptors in crowded areas that dissociate from scaffolds will face a longer escape time from the region and thus are more likely to rebind PSD-95.

If domains of high PSD-95 density tend to accumulate not only probes such as used here but also receptors, then their impact on synaptic transmission will depend on where they are with respect to sites of neurotransmitter release (MacGillavry et al., 2011). Recently, we have found that nanoclusters of PSD-95 frequently align transsynaptically with sites of neurotransmitter release, as indicated by their transsynaptic alignment with RIM1/2 molecules which in turn correlate with presynaptic vesicle fusion locations (Tang et al., in press). Thus, we speculate that the enhanced crowding within these high-density subdomains will slow and help limit the escape of receptors from points in the synapse where they are most likely to be activated during neurotransmission.

Crowding within high-density subdomains is likely not due only to postsynaptic scaffolding proteins. Indeed, transmembrane adhesion molecules, which associate with one another across the synaptic cleft, will enhance crowding further in the extracellular, transmembrane, and intracellular domains. The distribution of these adhesion molecules may regulate the alignment between neurotransmitter receptors and release sites, though this is not known. The TM probes employed here share commonalities with some of the intercellular adhesion molecules in size (e.g., single-pass TM proteins) or the ability to bind PSD-95, yet synaptic adhesion proteins display quite divergent patterns of expression within the synapse. SynCAM is distributed in clusters surrounding the border of PSD-95 molecules (Perez de Arce et al., 2015), as are some members of the cadherin/catenin system (Uchida et al., 1996). On the other hand, neuroligin1 and LRRTM2 are inside the synapse and distributed in clusters reminiscent of PSD-95 (Chamma et al., 2016).

Surprisingly, LRRTM2, though smaller in both extracellular and intracellular length than neuroligin1, in fact diffuses slower than neuroligin1 (Chamma et al., 2016). Moreover, LRRTM2 is more compactly distributed and in the synaptic center than the larger neuroligin1, suggesting that LRRTM2 is more likely associated with dense regions of PSD-95 which are often found near the center of the PSD (MacGillavry et al., 2013; Tang et al., in press). This paradoxical result confirms that the arrangements of TM proteins cannot be predicted by their bulk alone. Interestingly, however, AMPARs associate with LRRTM2 through their extracellular domains (de Wit et al., 2009), potentially regulating the mobility of each protein by the addition of further bulk and protein-protein interactions. Further experiments are needed to tease out how the various interactions on diverse synaptic TM protein species dictate one another's spatial arrangement.

The diffusion of TM proteins appeared complicated outside of the synapse particularly within few hundred nanometers from the border of the PSD. In some cases, the nonbinding probe moves over a large area in the extrasynaptic regions, but in other cases, they can appear confined or immobilized as the binding probes. Proteins other than scaffolding proteins can certainly affect the mobility of these probes outside the synapse. Some factors could affect both probes roughly equally: regions with high density of endocytic adaptor molecules (Blanpied et al., 2002; Petralia et al., 2003; Racz et al., 2004), zones of dense cortical cytoskeleton (He et al., 2016), sites of plasma membrane-ER apposition (Spacek and Harris, 1997). In addition, puncta adherens or clusters of adhesion molecules (Perez de Arce et al., 2015) may slow transit of all TM proteins. Less frequently, undetected regions could exist that would selectively affect the binding probe. For example, small and low-density regions below the PSD detection criteria (<10 molecules and <30 nm in diameter) could be situated within nanometers of the detected PSD border. These could come from small segments of multisegmented PSDs (Spacek and Hartmann, 1983; Stewart et al., 2005), but these types of PSDs are rare. In addition, though the density of PSD-95 outside the PSD border is quite low (Zhang and Diamond, 2009; Perez de Arce et al., 2015), its extrasynaptic mobility has not been measured and there may be enough PSD-95 in the perisynaptic region to bind and immobilize the binding probes. We speculate that the overall extrasynaptic diffusion of both probes are not different because few extrasynaptic regions selectively affect the binding probe but not the nonbinding probe.

### Role of Crowding in Synaptic Plasticity

The many types and time scales of ongoing and triggered PSD plasticity that have been documented (Okabe et al., 1999; Blanpied et al., 2008; MacGillavry et al., 2013; Bosch et al., 2014) suggest that PSD reorganization during plasticity will affect accumulation not just of TM proteins like AMPARs but additional molecules contributing to crowding as well. For instance, PSD-95 content in spines has been shown to decrease transiently after an LTP induction protocol in hippocampal slice cultures (Steiner et al., 2008). This may directly lead to loss of AMPARs as their binding partners are lost. However, the loss of PSD-95 may decrease crowding, which could prompt a net loss of even TM proteins with minimal direct binding to PSD-95 (e.g., desensitized AMPARs uncoupled from stargazin Constals et al., 2015). This further loss of molecular crowders could then further facilitate receptor exit. However, whether the transient loss of spine PSD-95 reflects a disruption of high-density areas of the synapse is unknown.

On the other hand, speculating further, if the transient decrease in synaptic PSD-95 during LTP induction reflects primarily loss from the PSD edge, and thus is not correlated with significant de-crowding at high-density regions, then the continued presence of additional nonbinding proteins at these high-density regions could in fact obstruct the entry of AMPARs to these regions and limit the changes in AMPARs level, at least over certain kinetic phases. Thus, it would be tempting to speculate in this case that the nonbinding proteins could affect LTP induction kinetics but not maintenance, as morphing dynamics and internal mixing of the PSD would eventually enable synapses to reach their new steady state capacity of AMPARs on a time scale of minutes. Resolving these many possibilities in the future will require close examination of the kinetics of protein redistribution and exchange during plasticity.

### Advantages and Disadvantages of PALM-PAINT

PALM of the PSD border improves discrimination of those molecules definitively within the synapse proper. However, we suspect that the effect of crowding may have been underestimated in our analysis because spatial and temporal alignment of the uPAINT and PALM data was subject to residual errors that may have diminished a larger underlying effect. The two color channels faced an alignment error of ∼6 nm, which would somewhat blur our measurement of regional PSD-95 density around individual tracked locations. In addition, the uPAINT data is subject to error stemming from the finite precision of individual localizations. The Atto647N we used for tracking is a relatively bright organic dye and helps to maximize this precision and thus minimize error in the estimate of Deff. However, brighter, longer-lasting fluorophores could be advantageous. Nanobody-labeled small quantum dots (Wang et al., 2014) have been used to track AMPARs in and around synapses, and have the additional advantage of being so bright as to facilitate tracking in 3D (Cai et al., 2014). However, 3D mapping of the PSD would require high localization numbers and longer imaging durations (Legant et al., 2016; and see below), and the z resolution normally obtainable without 4pi detection is usually worse than 100 nm for fluorescent proteins, making this difficult to implement.

In our application of PALM-PAINT, there was only limited temporal relationship between individual tracks (generally lasting <1 s and the PALM map (aggregated over the imaging session of generally 4–6 min)). Though lateral drift was corrected during this time (to an error we estimated as <10 nm), ongoing morphing and internal reorganization of the PSD (Kerr and Blanpied, 2012; MacGillavry et al., 2013) presumably degraded many details of the PSD-95 distribution in our final images. The reduced precision in capturing the true regional density of PSD-95 molecules would diminish the difference we saw between probes in different regional densities, and also reduce the difference between binding and nonbinding probe. Further, probes in a similar subsynaptic space but tracked early vs. late in the mapping might have not truly experienced the same degree of steric hindrance. However, the differences we observed were robust even in the face of these errors. Ideally, to capture the true effect size, one would need to monitor lateral drift continuously (Bon et al., 2015) and achieve more rapid mapping (Huang et al., 2013). However, in structures with low protein copy number, a large fraction of the proteins must be mapped to achieve statistical reliability (MacGillavry et al., 2013; Legant et al., 2016), which may preclude time-lapse imaging except if the protein exchange rate is high compared to the photobleaching rate induced by imaging.

We hope the combined approach of PALM-PAINT will help answer many key questions regarding synapse architecture and plasticity. One key issue is what mechanisms assemble the particular organization of PSD-95, a pattern that appears to dictate receptor number and position (Opazo et al., 2012). One possibility is that the more deeply positioned multi-domain proteins in the PSD, such as the Shank and GKAP families (Valtschanoff and Weinberg, 2001; Dani et al., 2010), may establish a platform of loose spacing with which the more superficial proteins such as PSD-95 may interact (Chen et al., 2008). Interestingly, in this case, a close interaction of the deeper PSD with cytoskeleton (Frost et al., 2010; MacGillavry et al., 2016) may thus provide a link between activity-dependent plasticity of spine and PSD structure. Alternatively, cleft-resident adhesion molecules have distinct organizational patterns (Perez de Arce et al., 2015; Chamma et al., 2016), that may guide intracellular protein organization in both the presynaptic and postsynaptic cells. Dissecting these possibilities, which require nanoscale resolution of position and mobility of multiple proteins, may be aided by future PALM-PAINT applications.

### AUTHOR CONTRIBUTIONS

TPL and TAB designed research; TPL performed research and data analysis; TPL and TAB wrote the article.

### REFERENCES


### FUNDING

This work was supported by the National Institutes of Health Grant F30MH102891 to TPL and R01MH080046 and R01MH096376 to TAB and the Kahlert Foundation to TAB.

### ACKNOWLEDGMENTS

We thank Sridhar Raghavachari and members of the Thompson, Jurado, and Blanpied laboratories helpful discussion, and Minerva Contreras for help with cultures.


adhesive nanodomains. Neuron 88, 1165–1172. doi: 10.1016/j.neuron.2015. 11.011


postsynaptic receptor. Proc. Natl. Acad. Sci. U S A 94, 6983–6988. doi: 10. 1073/pnas.94.13.6983

Zhang, J., and Diamond, J. S. (2009). Subunit- and pathway-specific localization of NMDA receptors and scaffolding proteins at ganglion cell synapses in rat retina. J. Neurosci. 29, 4274–4286. doi: 10.1523/JNEUROSCI.5602- 08.2009

**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Li and Blanpied. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Emergence of NMDA Receptor Metabotropic Function: Insights from Imaging

Kim Dore<sup>1</sup> , Jonathan Aow<sup>2</sup> and Roberto Malinow<sup>1</sup> \*

<sup>1</sup> Center for Neural Circuits and Behavior, Department of Neuroscience and Section for Neurobiology, Division of Biology, University of California at San Diego, San Diego, CA, USA, <sup>2</sup> Genome Institute of Singapore, Singapore, Singapore

The NMDA receptor (R) participates in many important physiological and pathological processes. For example, its activation is required for both long-term potentiation (LTP) and long-term depression (LTD) of synaptic transmission, cellular models of learning and memory. Furthermore, it may play a role in the actions of amyloid-beta on synapses as well as in the signaling leading to cell death following stroke. Until recently, these processes were thought to be mediated by ion-flux through the receptor. Using a combination of imaging and electrophysiological approaches, ion-flux independent functions of the NMDAR were recently examined. In this review, we will discuss the role of metabotropic NMDAR function in LTD and synaptic dysfunction.

Keywords: ion-flux independent, FRET-FLIM, long-term depression (LTD), NMDAR interactions with CaMKII, amyloid-beta induced depression, PP1, excitotoxicity

### INTRODUCTION

#### Edited by:

George Augustine, Nanyang Technological University, Singapore

### Reviewed by:

Kei Cho, University of Bristol, UK William N. Green, University of Chicago, USA

#### \*Correspondence:

Roberto Malinow rmalinow@ucsd.edu

Received: 25 April 2016 Accepted: 06 July 2016 Published: 28 July 2016

#### Citation:

Dore K, Aow J and Malinow R (2016) The Emergence of NMDA Receptor Metabotropic Function: Insights from Imaging. Front. Synaptic Neurosci. 8:20. doi: 10.3389/fnsyn.2016.00020 Transmembrane receptors have traditionally been divided into two classes: ionotropic and metabotropic. Ionotropic glutamate receptors (iGluRs) form channels that allow the passage of ions into the cell to drive signaling, while metabotropic glutamate receptors (mGluRs) generate downstream effects without ion-flux. The boundary between these two classes is not completely distinct, as there has been evidence that several iGluRs are capable of producing effects in the absence of ion-flux. For example, the N-terminal domain of GluA2, a subunit of the AMPA receptor (AMPAR), is sufficient to promote spine formation in hippocampal neurons (Passafaro et al., 2003). Another iGluR, the kainate receptor, can modulate GABA transmission without ion-flux (Rodríguez-Moreno and Lerma, 1998).

The NMDA receptor (NMDAR), a member of the iGluR family, is ubiquitously expressed and plays numerous roles in the brain (Traynelis et al., 2010). Given its ability to conduct calcium ions (Ca2+) well, it has been assumed that downstream signaling triggered by NMDARs was mediated by Ca2<sup>+</sup> influx and increased cytoplasmic Ca2+. However, to allow Ca2<sup>+</sup> entry through the receptor, several conditions have to be fulfilled: (1) glutamate must bind to GluN2 subunits; (2) glycine, the co-agonist must bind to GluN1 subunits; and (3) neurons must be sufficiently depolarized to eliminate the voltage-dependent magnesium ion (Mg2+) block of the channel. During high-frequency stimulation (HFS) these three conditions are met resulting in long-term potentiation (LTP; Bliss and Collingridge, 1993). For long-term depression (LTD), however, the role of the NMDAR is not as clear. A long-standing model has proposed that while LTP requires a large increase in cytoplasmic Ca2+, a moderate rise in cytoplasmic Ca2<sup>+</sup> would produce LTD (Lisman, 1989; Malenka, 1994). However, several recent studies indicate that NMDARs can induce LTD without ion-flow through the receptor (Nabavi et al., 2013; Dore et al., 2015; Stein et al., 2015; Carter and Jahr, 2016). Other publications have shown that excitotoxicity as well as amyloid-beta-induced synaptic depression depend on NMDAR activity but are likewise independent of ion-flow (Kessels et al., 2013; Tamburri et al., 2013; Birnbaum et al., 2015; Weilinger et al., 2016). In this review, we will discuss how these studies probed NMDAR metabotropic activity with an emphasis on the imaging techniques used.

### NMDAR-DEPENDENT LTD CAN BE INDUCED INDEPENDENTLY OF ION-FLUX

Interestingly, evidence for ion-flux independent LTD can be observed in data from older literature. Over 20 years ago, data were published indicating that MK-801, which blocks NMDAR channels, blocked LTP but failed to block LTD (Mayford et al., 1995). A similar effect was obtained by a different group (Scanziani et al., 1996). Surprisingly, these observations were not discussed in either study. The ion-flux dependence of LTD was recently examined more closely (Nabavi et al., 2013). Low-frequency stimulation (LFS) produced LTD in the presence of either MK-801 or 7-chloro-kynurenate (7CK, a competitive GluN1 antagonist; see **Figure 1A**) but not APV (a competitive GluN2 antagonist); all three antagonists effectively blocked ion-flux through the NMDAR. Moreover, LTD was observed in experiments in which intracellular Ca2<sup>+</sup> was clamped to basal levels, suggesting that a rise in intracellular Ca2<sup>+</sup> is not required for LTD. It was thus proposed that glutamate binding to the NMDAR could induce a conformational change in the cytoplasmic domain of the NMDAR that triggers downstream signaling resulting in LTD.

To test if ligand binding could drive movement of the NMDAR intracellular domain, FRET-FLIM [Forster resonance energy transfer measured by fluorescence lifetime imaging of the FRET donor, see **Toolbox** and (Wallrabe and Periasamy, 2005; Yasuda, 2006)] was employed (Dore et al., 2015). Recombinant GluN1 subunits of the NMDAR were tagged with GFP or mCherry at their carboxyl(c)-terminus and co-expressed in neurons. As the magnitude of FRET is very sensitive to the distance and orientation of the interacting fluorophores, nanometer-scale changes in distance can be reliably detected. Bath application or uncaging of glutamate in the presence of MK-801 or 7CK, but not APV, produced a transient change in FRET consistent with conformational movement of the NMDAR cytoplasmic domain (**Figure 1B**). Infusing neurons with a GluN1 c-terminus antibody through a patch pipette blocked the ligand-driven FRET change as well as LTD, suggesting that this conformational change is required for LTD induction.

Downstream signaling events were also examined using FRET-FLIM (Aow et al., 2015). Protein-phosphatase 1 (PP1) is one of the first molecules whose activity was shown to be required for LTD (Mulkey et al., 1993) and it co-immunoprecipitates with the NMDAR complex (Husi et al., 2000). FRET between GluN1-GFP and PP1-mCherry, observed in baseline conditions, was transiently reduced during chemical LTD induction. This ligand-driven decrease in FRET required NMDAR conformational movement but not PP1 activity (Aow et al., 2015). It is possible that the transient movement of PP1 relative to the NMDAR cytoplasmic domain exposes the catalytic active site of PP1 to a target unavailable under basal conditions. One potential target is calcium-calmodulin dependent protein kinase II (CaMKII; Strack et al., 1997), which is recruited to the NMDAR complex during LTP stimuli (Otmakhov et al., 2004) and whose activity is required for both LTP (Malenka et al., 1989; Malinow et al., 1989; Silva et al., 1992) and LTD (Coultrap et al., 2014). By monitoring FRET between fluorescently-tagged GluN1 and CaMKII, a delayed decrease in the NMDAR-CaMKII interaction was observed during ionflux independent LTD (Aow et al., 2015). This effect depended on PP1 activity and was not seen with a CaMKII mutant that cannot be dephosphorylated at Thr-286 (CaMKII-Thr-286-Asp), suggesting that dephosphorylation of Thr-286 is necessary to modify the NMDAR-CaMKII interaction (Aow et al., 2015). Co-immunoprecipitation experiments additionally revealed that the amount of total CaMKII bound to the NMDAR was unaffected by ion-flow independent LTD, whereas levels of phosphorylated Thr-286 were reduced both during and after LTD induction (Aow et al., 2015). These results are consistent with a model for ion-flux independent LTD in which glutamate binding to the NMDAR induces a conformational change in the NMDAR intracellular domain that facilitates PP1 access to and dephosphorylation of CaMKII at Thr-286, thereby repositioning the CaMKII holoenzyme within the NMDAR complex. The relocated CaMKII could in turn potentially act on a novel site of the GluA1 subunit (Ser-567) that undergoes phosphorylation during LTD (Coultrap et al., 2014). Consistent with this model, CaMKII phosphorylation of GluA1-Ser-567 does not require Ca2<sup>+</sup> or calmodulin (Coultrap et al., 2014). Ultimately, this process could increase AMPAR endocytosis (Lüscher et al., 1999; Lin et al., 2000; Kim et al., 2001; Shi et al., 2001) and lead to depressed synaptic transmission.

Ion-flow independent NMDAR activation of downstream signaling pathways has also been linked to shrinkage of dendritic spines (Stein et al., 2015). Stein et al. used 2 photon laser scanning microscopy (TPLSM) to monitor structural changes in the dendritic spines of GFPexpressing hippocampal neurons. Low-frequency uncaging of glutamate produced a ∼20% decrease in spine size that was independent of ion-flow through the NMDAR (**Figure 1C**). While high-frequency glutamate uncaging produced an increase in spine volume, the same stimulus in the presence of either 7CK or MK-801 led to spine shrinkage. This result is consistent with the finding that HFS (delivered electrically) in the presence of MK-801 produces LTD instead of LTP (Nabavi et al., 2013). Spine shrinkage was also abolished when p38 MAPK activity was blocked (Stein et al., 2015), which is again consistent with the observation that levels of phosphorylated p38 (which is the active form) were increased during ion-flow independent LTD (Nabavi et al., 2013). In the future, it will be important to elucidate how the initial movement in the NMDAR cytoplasmic domain subsequently affects signaling molecules, such as cofilin, calcineurin and p38, implicated in LTD.

An ion-flux-independent mechanism for NMDARdependent LTD has been challenged by some recent studies (Babiec et al., 2014; Volianskis et al., 2015; Sanderson et al., 2016) but confirmed by others (Kim et al., 2015; Stein et al., 2015; Carter and Jahr, 2016). It is notable that the experimental conditions used in these recent studies that failed to detect ion-flux-independent LTD were not identical to those supporting this form of LTD. For instance, NMDAR antagonists were typically acutely washed in and then out of the preparation during LTD induction (Volianskis et al., 2015), instead of being present throughout the experimental duration. Furthermore, control pathways, which monitor transmission onto the same neurons but do not receive the conditioning stimulus, were generally not included (Babiec et al., 2014; Volianskis et al., 2015; Sanderson et al., 2016). These differences in methodology are significant and can make an impact in the outcome and interpretation of results (Nabavi et al., 2014). Therefore, it will be important to compare carefully the experimental conditions employed by different studies. Nevertheless, it remains possible that two different, independent

#### TOOLBOX | FRET measurements using fluorescence lifetime imaging microscopy.

FRET is a non-radiative energy transfer mechanism between two fluorescent molecules. There are two main requirements for successful FRET. First the fluorescence emission of the FRET donor must overlap with the FRET acceptor absorption spectrum; and second, these fluorescent molecules must be no more than ∼10 nm apart from each other (Lakowicz, 2006). This spatial requirement of FRET is very sensitive; it can thus be used as a "molecular ruler" to assess protein structure (Gustiananda et al., 2004) or to monitor subtle conformational changes (Dore et al., 2015).

FRET can be measured by acquiring a series of images in different combinations of excitation and emission channels or by photobleaching of the FRET acceptor. However, these methods are prone to errors and are generally not well suited for measurements in living cells expressing fluorescent proteins (Selvin, 2000; Yasuda, 2006; Piston and Kremers, 2007). Fluorescence lifetime is defined as the average time a molecule stays in its excited state before emitting a photon (A). Because lifetime is an intrinsic property of fluorophores, it is independent of experimental conditions such as concentration, excitation intensity and photobleaching (Lakowicz, 2006; Yasuda, 2006). Importantly, by adding an additional route for the donor fluorophore to return to ground state, the degree of FRET makes the fluorescence lifetime of the donor proportionally shorter (A,B). To measure fluorescence lifetimes, the most common approach is time correlated single photon counting (TCSPC; Becker et al., 2004) which calculates the time delay between the detection of fluorescence photons and laser excitation pulses (B). When TCSPC is combined with laser scanning microscopy, it is possible to obtain fluorescence lifetimes, and hence detect changes in donor-acceptor distances, at every pixel of an image (B,C).

FRET-FLIM. (A) Jablonski diagram of FRET donor and acceptor energy levels. After excitation by a 1-photon (blue arrow) or 2-photon (brown arrows) laser, the FRET donor can return to ground state by emitting a photon (green arrow) or by transferring its energy to a nearby acceptor (dashed green arrows). (B) The fluorescence lifetime of the FRET donor, which becomes shorter with increased proximity of the FRET acceptor, is calculated with the fluorescence decay curve. (C) Time-correlated single photon counting (TCSPC) records fluorescence decay curves for each pixel of an image. By fitting these curves, the fluorescence lifetime of the FRET donor can be assessed. The FLIM image is then color coded according to the FRET donor lifetime at each pixel.

ion-flux through the NMDAR is not required for spine shrinkage. Modified from Stein et al. (2015).

forms of NMDAR-dependent LTD exist: one that requires ion-flow through NMDARs and one that does not. Different experimental conditions could selectively recruit either of these two forms.

### TRAFFICKING OF NMDAR IS REGULATED BY SYNAPTIC ACTIVITY BUT NOT ION-FLUX

In addition to its more recently described role in LTD, a few older studies have indicated that ligand binding to the NMDAR, in the absence of ion-flux, could control NMDAR trafficking (Vissel et al., 2001; Barria and Malinow, 2002; Nong et al., 2003). The Westbrook lab showed that even with its pore blocked, ligand binding to the NMDAR drove tyrosine dephosphorylation of GluN2A subunits, resulting in NMDAR endocytosis and decreased NMDA currents. Another group separately observed that an initial application of glycine was sufficient to prime NMDARs for subsequent use-dependent endocytosis, again leading to a decline in NMDA currents (Nong et al., 2003). Moreover, synaptic replacement of GluN2B- with GluN2A-containing NMDARs, an important developmentally controlled process (Hestrin, 1992; Monyer et al., 1994; Sheng et al., 1994; Stocca and Vicini, 1998; Tovar and Westbrook, 1999), required ligand binding without ion flux (Barria and Malinow, 2002). Interestingly, the replacement of synaptic GluN2B-containing NMDARs by newly synthetized GluN2B-containing NMDARs did not require ligand binding. It is important to note that these effects on NMDAR trafficking required both agonist and co-agonist binding to NMDARs as they were blocked by antagonists to the glutamate binding site on GluN2 subunits or to the glycine binding site on GluN1 subunits (Vissel et al., 2001; Barria and Malinow, 2002; Nong et al., 2003); in contrast, LTD only requires ligand binding to GluN2 subunits (Nabavi et al., 2013).

(AMPAR) currents of such a recording (top right). Bottom, results from paired-recordings performed in ACSF containing no drug (N = 41), Ro 25-6981 (N = 37) or MK-801 (N = 25); ∗∗indicates p < 0.001. Modified from Kessels et al. (2013). (B) Spine density is reduced in a transgenic mouse model of Alzheimer's disease (AD), APV blocks the effect but not memantine. Scale bar, 5 µm. Modified from Birnbaum et al. (2015). (C) Sustained NMDA application causes excito-toxic insults to CA1 neurons in the form of blebbing of dendrites (left panels). This effect is blocked by APV and CGP-78608 (middle panels) but not MK-801 (right panels). Scale bar, 25 µm. Modified from Weilinger et al. (2016).

### METABOTROPIC NMDAR ACTIVITY CAN INDUCE SYNAPTIC DYSFUNCTION

Recent results have suggested a role for metabotropic NMDAR activity in amyloid-beta mediated synaptic dysfunction, which may contribute to hippocampal deficits in Alzheimer's disease (AD) and precede neurological symptoms by a decade or more (Terry et al., 1991; Reiman et al., 1996). A number of studies using electrophysiology and imaging have reported that amyloid-beta impairs LTP, depresses synaptic transmission and induces synapse loss in various hippocampal preparations (Chapman et al., 1999; Larson et al., 1999; Walsh et al., 2002; Wang et al., 2002; Kamenetz et al., 2003; Snyder et al., 2005; Hsieh et al., 2006; Shankar et al., 2007; Wei et al., 2010; Birnbaum et al., 2015). The effect of intracellularly delivered amyloid-beta is not clear, as one report indicated synaptic depression (Ripoli et al., 2014) while another indicated synaptic potentiation (Whitcomb et al., 2015). In many of these studies the electrophysiological results could be corroborated using imaging. For instance, Wei et al. (2010) used TPLSM to show that GFP-filled dendritic spines close to axons or dendrites overexpressing amyloid-beta displayed a smaller increase in spine volume following a chemically-induced LTP protocol as compared to more distant spines, suggesting that secreted amyloid-beta impaired LTP. Hsieh et al. (2006) used TPLSM of AMPARs tagged with the pH-sensitive GFP-variant SEP (Super-Ecliptic-Phluorin) to measure surface AMPARs and found that amyloid-beta reduced surface GluA1 and GluA2. Likewise, immunostaining and imaging primary cultures treated with amyloid-beta revealed a reduction in surface NMDARs (Snyder et al., 2005) and AMPARs (Almeida et al., 2005; Alfonso et al., 2014). The decrease in synaptic AMPAR and NMDAR currents correlates, therefore, with a decrease in surface receptors as determined with optical techniques. Finally, several groups have shown using TPLSM or confocal microscopy that endogenously expressed or exogenously applied amyloid-beta reduces spine density in GFP-expressing neurons (Hsieh et al., 2006; Shrestha et al., 2006; Calabrese et al., 2007; Shankar et al., 2007; Wei et al., 2010; Zempel et al., 2010), which may explain the electrophysiologically observed decreased frequency of miniature excitatory postsynaptic currents (Kamenetz et al., 2003; Hsieh et al., 2006). Therefore, amyloid-beta induces synaptic insults that can be directly observed through imaging.

A mechanism proposed to account for synaptic impairment by amyloid-beta is enhanced ionotropic glutamate receptor endocytosis. Indeed, there is evidence that inhibiting endocytic signaling pathways or overexpressing mutant endocytic-resistant receptors can ameliorate the reduction in AMPAR and/or NMDAR currents (Snyder et al., 2005; Hsieh et al., 2006; Knafo et al., 2016). Notably, Kessels et al. (2013) reported that despite block of ion flux, not all NMDAR antagonists prevented amyloidbeta-induced depression of AMPAR-mediated transmission. The GluN2 antagonists (R)-CPP, Ro25-6981, and ifenprodil afforded a complete block; whereas the GluN1 antagonist 7CK and the NMDAR pore blocker MK-801 had no effect (**Figure 2A**; Kessels et al., 2013). Thus, the block of depression correlated with actions on different NMDAR subunits rather than block of ion-flux. In another model, amyloid-beta oligomers exogenously applied to organotypic hippocampal slices acutely depressed AMPAR-mediated transmission in a manner that was dependent on synaptic stimulation and NMDAR activation but not NMDAR ion-flux (Tamburri et al., 2013). Both studies therefore suggest that amyloidbeta activates a metabotropic NMDAR signaling pathway that depresses synaptic transmission. The evidence that this pathway could then be involved in eventual spine loss comes from three other studies using imaging techniques. Two studies (Shankar et al., 2007, 2008) showed using TPLSM that (R)-CPP

prevented amyloid-beta-induced spine loss in GFP-expressing organotypic slice neurons. Birnbaum et al. (2015) subsequently demonstrated that the competitive GluN2 antagonist APV also blocked spine loss in transgenic AD mice (as well as in hippocampal slices incubated in amyloid-beta oligomers), whilst MK-801, memantine, another NMDAR pore blocker, and buffering postsynaptic calcium ions with BAPTA had no effect (**Figure 2B**). That study (Birnbaum et al., 2015) also showed an amyloid-beta-induced reduction in PSD-95 and synaptophysin levels that was blocked by APV but not by MK-801 or memantine. Moreover, they demonstrated that p38 MAPK phosphorylation was increased by amyloidbeta in a NMDAR ion-flux independent manner and that spine loss depended on p38 MAPK activity (Birnbaum et al., 2015), supporting a link, previously examined (Hsieh et al., 2006), between LTD and amyloid-beta-induced depression. Taken together, these imaging results are in agreement with electrophysiological experiments and support the hypothesis that amyloid-beta toxicity depresses synaptic transmission via metabotropic NMDAR signaling that results in eventual spine loss.

The role of NMDARs in mediating excitotoxicity has been extensively described (reviewed in Choi, 1992), and it has been widely suggested that excessive Ca2<sup>+</sup> influx through the receptor is responsible for inducing cell death (Choi, 1995; Tu et al., 2010). Interestingly, recent findings suggest that a metabotropic NMDAR ''signalsome''—involving the NMDAR, the pannexin-1 channel (Panx1) and src kinase—is capable of inducing cellular dysfunction in response to excessive NMDAR stimulation (Weilinger et al., 2016). TPLSM was used to image fluorescently labeled CA1 neurons in acute rat hippocampal slices treated with a high dose of NMDA. This protocol induced blebbing in the dendrites of CA1 neurons as well as mitochondrial dysfunction, an effect that was blocked by co-application of the GluN1 antagonist CGP-78608 and APV, but not by MK-801, indicating that ligand binding to the NMDAR was capable of damaging neuronal morphology independently of ion flux through the receptor (**Figure 2C**). Interestingly, Panx1 channels appear to be located almost exclusively at the PSD (Zoidl et al., 2007), which suggests that this metabotropic NMDAR ''signalsome'' is synaptic. Additional experiments demonstrated that ligand binding, but not NMDAR ion flux, was necessary for downstream activation of Panx1 (Thompson et al., 2006, 2008; Weilinger et al., 2012). Metabotropic NMDAR activity did not change the degree of interaction between GluN1 and Panx1, but it did increase Src kinase binding to GluN1, Src activation, as well as Src-dependent phosphorylation

presence, spine elimination will occur.

of Panx1. Peptides that either disrupted the GluN1-Src interaction (Src48) or interfered with Panx1 phosphorylation (Tat-Panx308) were neuroprotective in vitro, and injection of Tat-Panx<sup>308</sup> reduced brain lesion volume in an in vivo model of stroke. Indeed, as ischemia in the brain is believed to drive subsequent excitotoxicity, these results suggest that targeting Src or Panx1 in a clinical setting could be therapeutically effective.

### CONCLUDING REMARKS

A number of studies have provided evidence that physiological and pathological processes can be triggered by ligand binding to the NMDAR, without requiring flow of ions through its pore. It will be important to determine conditions that control whether an ionotropic or metabotropic NMDAR

## REFERENCES


mechanism is engaged during LTD. Ion-flux independent LTD appears to be mediated by a movement in the NMDAR cytoplasmic domain that affects its interactions with at least two signaling proteins, PP1 and CaMKII. Subsequent signaling, with increased p38 MAPK phosphorylation likely, leads to AMPAR removal, shrinkage of dendritic spines and depressed synaptic transmission (**Figure 3**). Interestingly, it seems that if the stimulus recruiting metabotropic NMDAR function is sustained, as in the contexts of amyloid-beta overproduction or excitotoxic conditions following ischemia, metabotropic NMDAR activity can also lead to synaptic and neuronal dysfunction.

### AUTHOR CONTRIBUTIONS

KD, JA and RM wrote the manuscript. KD designed figures.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Dore, Aow and Malinow. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Correlative Light Electron Microscopy: Connecting Synaptic Structure and Function

Isabell Begemann1,2 and Milos Galic1,2 \*

<sup>1</sup> DFG Cluster of Excellence 'Cells in Motion', (EXC 1003), University of Muenster, Muenster, Germany, <sup>2</sup> Institute of Medical Physics and Biophysics, University Hospital Münster, University of Muenster, Muenster, Germany

Many core paradigms of contemporary neuroscience are based on information obtained by electron or light microscopy. Intriguingly, these two imaging techniques are often viewed as complementary, yet separate entities. Recent technological advancements in microscopy techniques, labeling tools, and fixation or preparation procedures have fueled the development of a series of hybrid approaches that allow correlating functional fluorescence microscopy data and ultrastructural information from electron micrographs from a singular biological event. As correlative light electron microscopy (CLEM) approaches become increasingly accessible, long-standing neurobiological questions regarding structure-function relation are being revisited. In this review, we will survey what developments in electron and light microscopy have spurred the advent of correlative approaches, highlight the most relevant CLEM techniques that are currently available, and discuss its potential and limitations with respect to neuronal and synapse-specific applications.

### Edited by:

Marc Fivaz, Duke NUS Graduate Medical School, Singapore

#### Reviewed by:

Jeff Lichtman, Harvard University, USA Christian Wilms, Scientifica Ltd., UK

### \*Correspondence:

Milos Galic galic@uni-muenster.de

Received: 23 March 2016 Accepted: 12 August 2016 Published: 23 August 2016

#### Citation:

Begemann I and Galic M (2016) Correlative Light Electron Microscopy: Connecting Synaptic Structure and Function. Front. Synaptic Neurosci. 8:28. doi: 10.3389/fnsyn.2016.00028 Keywords: correlative light electron microscopy, CLEM, synapse, neuron, fluorescence microscopy, electron microscopy, TEM, SEM

### INTRODUCTION

Neurons transmit the majority of trans-cellular signals via synaptic contacts. To correctly and reliably respond to various stimuli (e.g., different input frequencies), each synapse hosts an elaborate machinery to regulate signal transmission in a context-dependent manner. On the molecular level, synaptic transmission in its most simple form relies on fusion of ∼40 nm large vesicles at the presynaptic site, diffusion of released neuro-transmitters across the 30 nm wide synaptic cleft, and subsequent ligand-specific activation of receptors at the postsynaptic site. Considering the small size of this signaling unit, it is not surprising that major advancements in understanding synaptic function have been closely correlated with progress in imaging technology. For instance, the notion that communication across synaptic clefts relies on chemical signals would not have been possible without detailed information from electron micrographs on synapse ultrastructure (Derobertis and Bennett, 1955; Pappas and Bennett, 1966) and vesicle dynamics (Heuser and Reese, 1973). Likewise, our understanding about the function of individual synaptic proteins has substantially advanced with the introduction of ion and protein markers (Shimomura et al., 1962; Tsien, 1981; Heim et al., 1994) for fluorescence light microscopy. Intriguingly, and despite the fact that today's neuroscience is building up on information obtained with electron and light microscopy, strategies to combine these two approaches only recently have started gaining traction.

The emergence of such dual approaches, termed correlative light electron microscopy (CLEM), from a sparsely known branch of imaging to center stage can be linked to a series of landmark papers from the nineties (Deerinck et al., 1994; Svitkina et al., 1995). Since then, fueled by advances in imaging and labeling techniques, publications using CLEM approaches have steadily been rising. In this review, we will revisit what developments have incited the advent of correlative imaging approaches, and highlight the most relevant dual microscopy techniques that are currently available. We begin by giving an overview of electron and light-based imaging methods that can be used for CLEM, followed by a section describing how individual approaches can be successfully combined to create correlative approaches. Next, we will reflect on technical concerns that need to be taken care of when designing dual imaging experiments, before closing with a discussion on current limitations and future challenges associated with CLEM.

### ELECTRON AND LIGHT-BASED TECHNIQUES USED FOR CLEM

CLEM describes a continuously growing number of procedures that allow merging electron and light-based images from the same object. Thus, at least in theory, each existing electron and light microscopy technique where ultrastructure remains intact could be paired to generate a CLEM image. To illustrate this combinatorial potential, and to expound existing limitations, we begin this section by discussing electron and light microscopy techniques suitable for correlative approaches, before proceeding to examples where CLEM was successful applied to study neuronal and synaptic function.

### Electron Microscopy–Visualizing the Ultrastructure

Electron micrographs in CLEM rely to a large extent on transmission electron microscopy (TEM) and scanning electron microscopy (SEM). TEM, which enables visualization of 50– 100 nm thick cross-sections of samples with a resolution of down to a few Angstrom (Pierce and Buseck, 1974; Ruska, 1987; Erni et al., 2009), was critically involved in the detailed characterization of synaptic structures (Gray, 1959; Pappas and Bennett, 1966; Fifkova and Delay, 1982; Landis and Reese, 1983; Watanabe et al., 2013, 2014), and has helped to advance our understanding on age and disease-dependent changes in synaptic properties (DeKosky and Scheff, 1990; Navlakha et al., 2013).

Contrary to TEM, where electron shadows are used to create the image, SEM-based strategies utilizes the interaction of electrons with molecules in the sample to recreate the image. One commonly applied SEM strategy, where secondary electrons are used to determine the surface topography of objects with a precision of ∼1 nm (Vernon-Parry, 2000), was central for investigating surface features such as the morphology of neuromuscular junctions (Desaki and Uehara, 1981) as well as structural reorganization during exploratory dendritic filopodia formation in hippocampal neurons (Galic et al., 2014). In addition, back-scattered electrons created by interaction with heavy elements (i.e., high atomic number) in the uncoated sample can also be used to render images comparable to transmission electron micrographs of ultrathin sections, an approach that is of particular relevance for embedded sections on an electronopaque surface or samples that are investigated 'en block' (Briggman et al., 2011).

A single cross-section, or the surface of the sample, is lacking information on the three-dimensional organization within the biological specimen. Here, cryo-fracture, where frozen samples are broken to expose cell structures along the fracture line (Castejon and Caraballo, 1980; Castejon, 1996), or unroofing, where intracellular structures are uncovered by brief bursts of ultrasound (Lang, 2003), can be used together with deep-etching (Heuser and Salpeter, 1979) to gain insights into subcellular organization. While useful, these strategies lack the capability for systematic three-dimensional sample reconstruction that can be achieved with serial sectioning, where successive ultrathin slices are imaged. Early attempts of serial sectioning reach back to the late 60s (De Rosier and Klug, 1968), and have since then continuously been used to investigate neuronal networks (White et al., 1986) and synaptic connection (Mishchenko et al., 2010). In modern serial section TEM (ssTEM), sections of 60 ± 20 nm are cut with an ultra-microtome and placed manually on a metal support grid (Harris et al., 2006). More recently, automated tapecollecting ultra-microtome SEM (ATUM-SEM) has emerged as a powerful alternative (Hayworth et al., 2006; Kasthuri et al., 2007, 2015; Terasaki et al., 2013; Morgan et al., 2016). Here, sections of 30 nm are cut by an ultra-microtome and collected from the water bath using a conveyor-belt like support tape. As the support tape is electron-opaque, a finely focused SEM beam is applied to raster the surface of the sample and backscattered electrons are used to reconstruct the image. Thus, compared to ssTEM, ATUM-based strategies not only allow thinner sectioning and rapid imaging of larger areas but also substantially reduce errors associated with manual sample handling (Hayworth et al., 2014). Alternatively, an embedded tissue block can also be sectioned directly within the SEM vacuum chamber, either using a diamond knife (serial block-face SEM, SBEM; Leighton, 1981; Denk and Horstmann, 2004) or by milling with a focused ion-beam (FIB-SEM; Watkins et al., 1986; Knott et al., 2008). In the latter, a scanning electron microscope is used to image the surface of the embedded sample, while a high current focused ion beam continuously slices off sections perpendicular to the SEM axis, thus allowing 3D reconstruction of the sample. Compared to diamond-based sectioning, FIB-SEM is thus not only faster but also permits sectioning samples with a step size in the single nm range (Knott et al., 2008; Merchan-Perez et al., 2009; Lehmann et al., 2014).

Another strategy for detailed three-dimensional reconstructions relies on electron tomography (ET). In this approach, a tilt series (usually ranging from −60◦ to +60◦ ) of two-dimensional images is generated to reconstruct the three-dimensional shape of an object within the single slice (Crowther et al., 1970; Hoppe et al., 1974; Penczek, 2010; Ercius et al., 2015). ET was successfully used to resolve ultrastructural features of the presynapse (Perkins et al., 2015) and synaptic vesicle populations in saccular hair cells (Lenzi et al., 1999). Yet,

while ET is suitable for atomic-resolution, radiation damage, and limitations in interpretability for thicker sections need to be considered when preparing the sample (Tocheva et al., 2010).

Finally, we would like to note that further electron-based imaging strategies exist (e.g., STEM (Crewe et al., 1970; Engel, 1978), tSEM (Kuwajima et al., 2013)). While not commonly used in CLEM studies, it is plausible to envision that some of these additional types of electron microscopy may be beneficial for a particular question. For readers interested in learning more on electron microscopy techniques, we recommend reading one of the many excellent reviews available on this exciting topic (Briggman and Bock, 2012; Knott and Genoud, 2013; Miranda et al., 2015; Perkins et al., 2015).

### Light Microscopy–Monitoring and Manipulating Cell Function

While the electron microscopy techniques described above are well suited to investigate ultrastructural features with high axial resolution, they lack spatio-temporal information available with light microscopy. To increase penetration depth and reduce background compared to whole field illumination (Epi), CLEM studies frequently rely on confocal microscopy, where a pinhole is used to reduce out-of-focus light (Minsky, 1988), light sheet microscopy, where a light beam perpendicular to the objective illuminates only the focal plane (Engelbrecht and Stelzer, 2006; Keller et al., 2008), or two-photon microscopy, where simultaneous absorption of two photons is used for spatially controlled illumination (Denk et al., 1990). Resolution in light and electron microscopy is based on numerical aperture and wavelength. However, unlike in electron microscopy, where resolution is limited by the spot size (SEM) or the grain of the detector (TEM), the limiting factors in light microscopy is the wavelength (Rayleigh, 1879; Abbe, 1883). To reach beyond the diffraction limit, several approaches have been introduced over the last decade. Among the most prominent super-resolution microscopy techniques used in CLEM studies are stimulated emission depletion microscopy (STED), where a depletion laser limits the width of the emitting light source (Hell and Wichmann, 1994; Klar et al., 2001), and stochastic techniques, such as fluorescent photo-activation localization microscopy (PALM; Betzig and Chichester, 1993; Betzig et al., 2006) or stochastic optical reconstruction microscopy (STORM; Rust et al., 2006), where light emitted from sequentially activated fluorophores is fitted to determine its precise localization. As before, we would like to note that additional light-based approaches [e.g., total internal reflection fluorescence microscopy (Axelrod, 1981) or structured illumination microscopy (Neil et al., 1997)] can be used in CLEM approaches. Readers interested to learn more about fluorescence microscopy techniques, we refer to one of the many excellent reviews focused on these topics (Lichtman and Conchello, 2005; Combs, 2010; Huang et al., 2010).

Fluorescence light microscopy not only allows to visualize cells within neuronal circuits (Livet et al., 2007), but a continuously growing number of molecular probes and genetically encoded markers also provide tools to study a variety of neuronal and synaptic properties. Parameters that can be analyzed include among others subcellular protein localization (Chalfie et al., 1994; Heim et al., 1994; Kilgore et al., 2013), protein activity (Adams et al., 1991), ion dynamics (Tsien, 1980, 1981; Tsien et al., 1982; Minta and Tsien, 1989; Nakai et al., 2001), pH (Tanasugarn et al., 1984; Miesenbock et al., 1998), membrane potential or voltage (Davila et al., 1973; Siegel and Isacoff, 1997; Zochowski et al., 2000), or lipid species (Stauffer et al., 1998). Intriguingly, light is also suitable to actively manipulate the biological sample in a precise spatio-temporal manner. One commonly used approach relies on light-activation of caged substrates. Relevant for neurobiology, strategies for uncaging calcium (Ellisdavies and Kaplan, 1994), IP3 (Wang and Augustine, 1995), and various neurotransmitters (Ellis-Davies, 2011) have been realized. More recently, light has also been used to regulate protein function (Cao et al., 2008; Tischer and Weiner, 2014). This approach, coined optogenetics, has provided among others tools for controlling neuronal and synaptic activity with unprecedented spatio-temporal precision (Boyden et al., 2005; Fenno et al., 2011; Rost et al., 2015). Finally, it is worth mentioning that even physical cell parameters, such as shape or membrane tension, can be altered by light in living samples, using for instance optical tweezers (Ashkin, 1970; Zhang and Liu, 2008). In summary, when combined with electron microscopy, these light-based approaches allow precisely altering a specific parameter while monitoring the cellular responses followed by analysis of the corresponding ultrastructural features.

## Correlative Light and Electron Microscopy in Neuroscience

Although the potential of combining ultrastructural information with functional studies was early noticed, first attempts to combine these microscopy techniques were limited to depicting separately prepared and imaged biological samples (Porter et al., 1945). Experiments where fluorescence and EM images of subcellular structures from the same cell were aligned started appearing in the 1970s (Hollander, 1970; Nakai and Iwashita, 1976), and have since then been applied to investigate a variety of neurobiological questions.

To fully understand neuronal or synaptic function, the cellular context needs to be considered. It is thus not surprising, that large efforts have been put into elucidating ultrastructural information within the intact tissue. In situ CLEM, where data from lifecell fluorescence imaging is combined with EM micrographs, has advanced our understanding of neuronal circuits (Bock et al., 2011; Briggman et al., 2011; Lee et al., 2016), and has also shed light on subcellular behaviors such as axosome shedding (**Figure 1A**) (Bishop et al., 2004). An alternative strategy, frequently applied to gain information of individual neurons in situ, relies on array tomography (AT), where plasticembedded tissue samples are sliced with an ultra-microtome, bonded array-wise onto a glass coverslip, stained, and finally imaged by fluorescence and electron microscopy (**Figure 1B**) (Micheva and Smith, 2007; Rah et al., 2013; Collman et al., 2015). By combining super-resolution techniques and serial sectioning, it was even possible to detect presynaptic dense projection proteins at a lateral resolution of 35–65 nm (Watanabe

et al., 2011). While serial sectioning-based CLEM approaches allow studying biological samples with high axial and lateral resolution, sample preparation, and 3D image alignment can be challenging (Micheva and Smith, 2007; Collman et al., 2015). Here, SBEM and FIB-SEM based approaches, where en-block EM imaging and fluorescence approaches are merged to obtain CLEM images, have helped to streamline in situ studies focused on the complexity of the nervous system (Briggman et al., 2011; Blazquez-Llorca et al., 2015), or to reconstruct whole cortical neurons (Maco et al., 2013) and single synapses (Bosch et al., 2015).

Orthogonal to in situ approaches described above, CLEM can also be applied to neurons isolated from brain tissues and cultured in vitro (Banker and Cowan, 1977). As such, cultured neurons are compatible with each CLEM technique described above (**Figure 1C**) (Al Jord et al., 2014; Paez-Segala et al., 2015; Fares et al., 2016). Although these cells lack the physiological context usually encountered by neurons within a tissue, cultured neurons may still yield advantages compared to in situ samples depending on the biological question. One such example, and complementing the section-based CLEM assays described so far, is the analysis of processes at the cellular surface, such as curvature-dependent protein recruitment to deforming plasma membranes (**Figure 1D**) (Galic et al., 2014).

### TECHNICAL CONSIDERATIONS

In this section, we will focus on technical possibilities and limitations that need to be taken into consideration when designing a CLEM experiment. We start by discussing challenges arising from sample preparation. Next, we will survey what markers are suitable for which technique, and discuss strategies for alignment of corresponding fluorescence images and electron micrographs with respect to their potential and drawbacks.

### Sample Preparation for CLEM Images

The quality of correlative image alignment critically relies on the ability to maintain the native organization of the cell during fixation and subsequent sample preparation. Thus, CLEM techniques not only need to be optimized for signal strength, but also for shape preservation. Formaldehyde, glutaraldehyde, and osmium tetroxide (OsO4), have been the standard fixatives for decades, including for brain tissues that are due to their softness easily damaged during fixation and subsequent processing. Classical fixation protocols include whole animal perfusion via the vascular system for larger specimens (e.g., whole brain) as well as immersion of tissue slice and cultured cells. Once the biological sample is fixed, the actual preparation for the respective EM technique is rendered. For slice-based assays, two different approaches exist: the pre-embedding method and the post-embedding method. For both methods, correlative approaches have been reported (Watanabe et al., 2011; Kopek et al., 2012). In the pre-embedding method, the antigen–antibody reaction is performed before (i.e., pre) plastic embedding and subsequent ultrathin sectioning. While better for preserving ultra-structures, this approach is often hindered by poor penetration of the antibody. In the post-embedding method, the antigen–antibody reaction is performed after (i.e., post) plastic embedding. As labeling takes place on thin tissue slices, antigens are more easily accessible. However, OsO4, which is often used as fixative and stain for membrane structures, can quench the fluorescence signal, and epoxy-based resins used due to its good preservation and ultra-sectioning properties partially inhibit photo-switching of fluorophores (Kim et al., 2015). Here, a number of modifications, such as using acrylic resins as embedding material or replacing OsO<sup>4</sup> with uranyl acetate, tannic acid or p-phenylenediamine, have helped to enhance compatibility and immunoreactivity (Phend et al., 1995; Osamura et al., 2000; Akagi et al., 2006; Kim et al., 2015). Complementing these slice-based assays, cells can also be prepared for analysis of surface features. As before, samples are fixed. However, rather than embedding the sample, a platinum replica of the surface is generated (Heuser et al., 1976), or the sample is subjected to critical point drying and surface staining, prior to image acquisition via SEM. Finally, one last concern is that chemical fixation is not instantaneous (Szczesny et al., 1996). To circumvent this issue, quick-freezing, where samples are 'slammed' on a super-cold block of metal sprayed with liquid helium (Heuser et al., 1976) or high-pressure freezing, where samples are frozen in milli seconds while pressure is increased to avoid water crystallization (Hunziker et al., 1984; Dahl and Staehelin, 1989), have improved preservation quality.

### Picking the Right Markers for CLEM

Unfortunately, not all markers are equally well suited for individual CLEM approaches. To account for these limitations, a series of labeling techniques has emerged. Considering its widespread availability, the most commonly used markers for CLEM are genetically encoded fluorescence tags, such as GFP, that can be detected in electron micrographs via immunogold labeling (**Figure 2A**) (Kukulski et al., 2011; Galic et al., 2012). In order to distinguish several proteins in the same sample, immuno-gold particles of variable sizes directed against different fluorescence tags can be used. However, since sample fixation may interfere with genetically encoded fluorescence markers (Muller-Reichert and Verkade, 2014) and gold particles may quench the fluorophore in parallel fluorescence/electron imaging (Kandela and Albrecht, 2007), additional markers are desirable. One such alternative relies on quantum dots (QDs; **Figure 2B**), which were shown to have a 10 times higher labeling efficiency than immuno-gold (Giepmans et al., 2005; Kuipers et al., 2015). These semiconductor nanocrystals consist of a cadmium selenide core surrounded by a zinc sulfide shell coated with affinity ligands (e.g., antibodies) for targeting the desired biomolecules (Alivisatos, 1996). Intriguingly, as the fluorescence wavelength of these photo-stable nanocrystals depends on their core size, it is possible to label samples with fluorescently different QDs, and then to distinguish individual QDs in electron micrographs by size (Giepmans et al., 2005; Kuipers et al., 2015). A minor disadvantage of QDs, however, lies in the difficulty to precisely measure sizes of small QDs due to its low contrast in EM (Brown and Verkade, 2010). While widely used, both

scanning electron microscopy (SEM) images of 70 nm section from the mouse cerebral cortex. From left to right: immunostaining of ultrathin sections for tubulin, GABA, SNAP-25, β-actin, and SEM image. Below, the boxed region is shown at a higher magnification. (C) In vitro CLEM of confocal and TEM images of hippocampal neurons. Ultrastructure of hα-Syn inclusions in cultured neurons from SNCA+/<sup>−</sup> mice shown in TEM (top), confocal image (middle), and as merged CLEM image (bottom). For illustration, the nucleus is rendered in blue, the cytosol in purple, and inclusions in yellow. To the right, high-resolution images of inclusions (red boxes 1–3), with arrowheads indicating filamentous structures, are shown. (D) In vitro CLEM of confocal and SEM images of cultured hippocampal neurons. Alignment of SEM and fluorescence signal for actin in cultured hippocampal neurons (top) and magnified sections of actin-rich convoluted nodes that form along dendritic arbors (bottom) are shown. Scale bars: (A) 25 µm at the top left, 1 µm at the top right, 0.25 µm at the bottom; (B) 10 µm at the top and 2 µm at the bottom; (C) 5 µm; (D) 2 µm. Pictures reprinted with permission from: (A) (Bishop et al., 2004); (B) (Micheva and Smith, 2007); (C) (Fares et al., 2016); (D) (Galic et al., 2014).

FIGURE 2 | Examples of markers commonly used for CLEM. (A) To the left, cartoon depicting immune-gold labeling of genetically encoded fluorescent protein. Note difference between the relative position of signal in fluorescence (yellow) and electron microscope (blue) images caused by antibodies. To the right, an example showing fluorescence and electron micrographs of HIV particle labeled with MA-EGFP on MDCK cells expressing RFP-tagged Histone 2B. (B) To the left, cartoon depicting protein and signal from QD in fluorescence (yellow) and electron microscope (blue) images. Note difference between protein epitope recognized by antibody and QD signal position. To the right, example depicting RFL6 fibroblasts fixed and stained with primary antibodies followed by secondary antibodies linked to QDs. QDs identify Cx43 at gap junctions and trafficking intermediates (green) and α-tubulin in microtubules (red). (C) To the left, cartoon depicting genetically encoded MiniSOG, as well as the relative position and signal shape for fluorescence (yellow) and electron microscope (blue) images. To the right, an example showing fluorescence and electron micrographs of HeLa cells expressing miniSOG labeled α-actinin. (D) To the left, cartoon depicting HaloTag labeling of protein as well as position and signal shape for fluorescence (yellow) and electron microscope (blue) images. To the right, an example showing fluorescence and electron micrographs of Hela cell transfected with Palmitoyl-HaloTag-meGFP. (E) Dendrites of medium-size spiny neurons in the rat neostriatum labeled with membrane-targeted GFP and immunolabeled with Cy5 against vesicular glutamate transporter2 (VGluT2; top). After detection by fluorescence microscopy, GFP and VGluT2 immunoreactivities were further developed for focused ion-beam SEM (FIB-SEM) via immunogold/silver enhancement and immunoperoxidase/DAB methods, respectively (bottom). Scale bars: (A) 5 µm to the left and 100 nm to the right; (B) 5 µm; (C) 2 µm to the left and 1 µm to the right; (D) 1 µm to the left and 500 nm to the right; (E) 3 µm. Pictures reprinted with permission from: (A) (Kukulski et al., 2011); (B) (Giepmans et al., 2005); (C) (Shu et al., 2011); (D) (Liss et al., 2015); (E) (Sonomura et al., 2013).

methods are hampered by a localization error due to spatial separation between the protein epitope and the electro-dense particle that is visualized in light and electron micrographs (Kandela et al., 2007; van Donselaar et al., 2007). To bypass this imprecision, Aptamers directed against the fluorescence tag (Shui et al., 2012) linked to gold (Javier et al., 2008; Chang et al., 2013), or nano-bodies directed against fluorescent proteins that can be conjugated to gold nanoparticles (Van de Broek et al., 2011; Ries et al., 2012) may provide promising alternatives.

Orthogonal to these approaches, several strategies have emerged that take advantage of singlet oxygen generators. Genetically encoded singlet oxygen generators are of particular relevance for samples that are investigated 'en block' (i.e., SBEM and FIB-SEM) and thus preclude post-staining of individual sections. One such example is the genetically encoded miniSOG (**Figure 2C**), which not only fluoresces when illuminated by blue light, but also yields an osmophilic reaction product by catalyzing the polymerization of 3,3<sup>0</sup> - Diaminobenzidine (DAB) into electron-dense polymers that are detectable in electron micrographs (Shu et al., 2011). The same strategy is also applicable using the Halo-Tag system, where a modified haloalkene dehalogenase (Los et al., 2008) is used to covalently bind to specific ligands such as tetramethylrhodamine (TMR; **Figure 2D**). Like miniSOG, TMR fluoresces upon illumination and is visible in EM due to DAB oxidation (Liss et al., 2015). Other self-labeling systems are provided by SNAP/CLIP-tags, in which the enzyme domain is bound to the protein of interest and labeled with a cell–permeable fluorescent ligand (Liss et al., 2015). Finally, cells can also be transfected with APEX, a non-fluorescent peroxidase that withstands strong fixation to yield light-independently EM contrast (Martell et al., 2012). However, considering that the reaction product can diffuse, these approaches may yield best results when staining enclosed structures. Notably, multiple procedures can be combined within a sample for visualization (**Figure 2E**).

### Strategies of Aligning Fluorescence Image and Electron Micrograph in CLEM

Correlation of fluorescence images and electron micrographs can be challenging for a number of reasons. For one, the region of interest can cover less than 0.0004% of the total sample area (Begemann et al., 2015), thus offering only on a limited number of reference points for navigation. And even after the region of interest acquired at the light microscope is found again on the electron microscope, rotation, magnification, and tilting angle still need to be adjusted for image alignment. This is of particular relevance in pre-embedding approaches, where, unlike to post-embedding strategies, the angle of the respective acquisition planes may differ. To tackle these obstacles, several strategies for navigation and image alignment have emerged for in situ and in vitro approaches.

For in situ CLEM, various fiduciary markers have been described that are equally suitable for pre-embedding and post-embedding strategies. In both approaches, the region of interest in the embedded sample is identified using a previously determined set of reference points. Here both, biological (e.g., blood vessels and dirt) and artificial (e.g., silica beads and gold markers) landmarks have proven to be useful for navigation. Once identified, individual EM sections are made and, if desired, 3D reconstruction of individual slices can be prepared by manual or automatic alignment using fiduciary landmarks present in adjacent sections (Bishop et al., 2004; Bock et al., 2011; Lee et al., 2016). Finally, electron micrographs (2D or 3D) are merged with the corresponding fluorescence images either manually or automatically using large features visible in both images (e.g., blood vessels and silica beads) for coarse alignment. For subsequent fine-adjustment, subcellular structures (e.g., gold fiduciary markers, nucleus, and filopodia) can be used. To estimate how accurate the alignment is, crosscorrelation analysis between light and electron micrographs can be performed.

In vitro CLEM approaches permit the generation of additional reference systems for image alignment on the plating substrate (i.e., glass and plastic). For example, a unique reference point can be created at the position where the fluorescence image is acquired, using for instance laser etching (Colombelli et al., 2008; Bishop et al., 2011; Urwyler et al., 2015) or a scratching device such as diamond objective markers (Sochacki et al., 2014). Upon processing, this reference point can then be used for navigation on the electron microscope and for subsequent image alignment. Another popular reference system relies on pre-formed structured (Svitkina, 2009; Benedetti et al., 2014) and stochastic (Begemann et al., 2015) micro-patterns or fiducial landmarks (Kukulski et al., 2011) for re-finding the region of interest on the EM as well as subsequent image alignment. Intriguingly, such reference points are not only suitable for manual alignment, but also for software-assisted solutions (Begemann et al., 2015), rendering these approaches an attractive entry point compared to more sophisticated CLEM strategies.

Finally, simultaneous fluorescence and electron imaging may provide an alternative strategy. Here, a number of custom-made (Liv et al., 2013; Nishiyama et al., 2014; Peddie et al., 2014; de Boer et al., 2015) and commercial (e.g., from Fei and Zeiss) instruments have been presented. These instruments, which are equally suitable for in situ and in vitro samples, integrate lightand electron imaging in one apparatus (iCLEM), thus ensuring dual imaging of the region of interest without need of later image alignment. However, as sample preparation for iCLEM is restricted to techniques suitable for parallel LM and EM (Agronskaia et al., 2008), and due to the limited availability of such instruments, iCLEM has not yet reached the optional users.

### CLEM–POTENTIAL AND LIMITATIONS

In this review, we have discussed possible combinations for correlating optical and electron microscopy approaches. We mentioned how improvements in imaging techniques, sample preparation, markers, and image alignment have facilitated

development of novel CLEM approaches, and showed examples where these strategies were successfully used for structurefunction analysis in neurons and synapses. In order to visualize the combinatorial potential, we provide a chart summarizing published CLEM pairings (**Figure 3**).

While correlative approaches have revealed aspects of neuronal and synaptic function that would not have been possible without it, challenges still remain due to the limited temporal and spatial resolution of CLEM. From this perspective, recent work on liquid cell electron microscopy (Ross, 2015), where electron-permeable membranes are used to host aqueous solutions under atmospheric pressure conditions for TEM and SEM imaging (Danilatos et al., 2011; Ominami et al., 2015), cryo-ET to study the supramolecular architecture of cellular structures (Medalia et al., 2002; Beck et al., 2007), and the continuous increase in sophistication and computational power used for image analysis have captured the imagination of people. While it remains elusive whether these and other emerging techniques will merge or replace CLEM-based studies, they are exemplary for an ongoing push toward visualization and comprehensive analysis of biological function on the structural level.

Taken together, the possibilities of applying correlative approaches to study neuronal and synapse function are growing thanks to the continuous progress in techniques, tools and protocols for combining these two types of microscopy. Alas, while correlative approaches in neurobiology have substantially increased in quality and availability, truly artifactfree preparation techniques for precise molecule-localization are yet to be achieved.

### AUTHOR CONTRIBUTIONS

fnsyn-08-00028 August 20, 2016 Time: 14:30 # 9

IB and MG conceived the idea, prepared illustrations for publication and wrote the manuscript.

### FUNDING

The authors acknowledge funding from the Cluster of Excellence 'Cells in Motion' (DFG, EXC-1003) to IB (PP-2014-04) and MG (FF-2016-03).

### REFERENCES


### ACKNOWLEDGMENT

The authors thank members of the Galic and Matis labs for critical reading of the manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnsyn. 2016.00028


for measuring membrane potential. Nat. New Biol. 241, 159–160. doi: 10.1038/newbio241159a0



visualized by cryoelectron tomography. Science 298, 1209–1213. doi: 10.1126/science.1076184



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Begemann and Galic. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

## Advanced Fluorescence Protein-Based Synapse-Detectors

Hojin Lee1,2† , Won Chan Oh1† , Jihye Seong2,3\* and Jinhyun Kim1,2 \*

<sup>1</sup> Center for Functional Connectomics, Korea Institute of Science and Technology, Seoul, South Korea, <sup>2</sup> Neuroscience Program, Korea University of Science and Technology, Daejeon, South Korea, <sup>3</sup> Center for Diagnosis Treatment Care of Dementia, Korea Institute of Science and Technology, Seoul, South Korea

The complex information-processing capabilities of the central nervous system emerge from intricate patterns of synaptic input-output relationships among various neuronal circuit components. Understanding these capabilities thus requires a precise description of the individual synapses that comprise neural networks. Recent advances in fluorescent protein engineering, along with developments in light-favoring tissue clearing and optical imaging techniques, have rendered light microscopy (LM) a potent candidate for large-scale analyses of synapses, their properties, and their connectivity. Optically imaging newly engineered fluorescent proteins (FPs) tagged to synaptic proteins or microstructures enables the efficient, fine-resolution illumination of synaptic anatomy and function in large neural circuits. Here we review the latest progress in fluorescent protein-based molecular tools for imaging individual synapses and synaptic connectivity. We also identify associated technologies in gene delivery, tissue processing, and computational image analysis that will play a crucial role in bridging the gap between synapse- and system-level neuroscience.

#### Edited by:

George Augustine, National Technical University, Singapore

#### Reviewed by:

Lucas Pozzo-Miller, The University of Alabama at Birmingham, USA Maurizio Giustetto, University of Torino, Italy

### \*Correspondence:

Jihye Seong jseong@kist.re.kr Jinhyun Kim jinnykim@me.com

†Co-first author.

Received: 09 May 2016 Accepted: 13 June 2016 Published: 30 June 2016

#### Citation:

Lee H, Oh WC, Seong J and Kim J (2016) Advanced Fluorescence Protein-Based Synapse-Detectors. Front. Synaptic Neurosci. 8:16. doi: 10.3389/fnsyn.2016.00016 Keywords: fluorescent protein sensors, synaptic connectivity, synapses, gene delivery, mapping and localization, light microscopy

### INTRODUCTION

The synapse is the primary site for neurons to make functional contacts for exchanging information. The term ''synapse'', meaning conjunction in Greek (synapsis = together + to fasten), was coined in 1897 by the eminent physiologist Charles Scott Sherrington (Nobel Laureate 1932). But the idea that synapses play critical roles as dynamically polarized, communication contacts was proposed

**Abbreviations:** 3 <sup>0</sup>UTR, 3<sup>0</sup> untranslated region; AMPAR, α-amino-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; CA3, Cornu Ammonis 3; CALI, chromophore-assisted light inactivation; CaMKII, calcium/calmodulindependent kinase II; CFP, Cyanfluorescent protein; CD4, cluster of differentiation 4; ChR, Channelrhodopsin; CRISPR, clustered regularly-interspaced short palindromic repeats; ENABLED, endogenous labeling via exon duplication; EM, electron microscopy; FMN, flavin mononucleotide; FP, fluorescent protein; fSPIM, fluorescent selective plane illumination microscopy; GFP, green fluorescent protein; HA, hemagglutinin; HSV-1, Herpes simplex virus-1; iDISCO, immunolabeling-enabled three-dimensional imaging of solvent-cleared organs; ID-PRIME, Interaction-Dependent Probe Incorporation Mediated by Enzymes; InSynC, Inhibition of Synaptic Release with CALI; KI, knock-in; LAP, lipoic acid acceptor peptide; LM, light microscopy; LOV2, light, oxygen, and voltage 2; LpIA, lipoic acid ligase; mGluR, metabotropicglutamate receptors;mGRASP,MammalianGFPReconstitutionAcrossSynapticPartners;miniSOG,mini small singlet oxygen generator; NCBI, National Center for Biotechnology Information; NS3, Nonstructural protein 3; OFP, Orangefluorescent protein; PDZ, PSD-95, Drosophila disc large tumor suppressor (Dlg1), and zonula occludens-1 protein (zo-1); PSD-95, postsynaptic density protein-95; rAAV, recombinant adeno-associated virus; SM protein, Sec1/Munc18-like protein; SNARE, SNAP (Soluble N-ethylmaleimide-sensitive factor attachment protein) Receptor; syb, synaptobrevin; spGFP, split-GFP; sypHTomato, Synaptophysin-fused pHTomato; TimeSTAMP, Time-Specific Tagging for the Age Measurement of Proteins; VAMP2, vesicle-associated membrane protein 2; VenusNT, Venus N-terminal; VenusCT, Venus C-terminal; VGluT, vesicular glutamate transporter; YFP, Yellow fluorescent protein.

by Santiago Ramon y Cajal (Nobel Laureate 1906). Since then, the synapse as a structural and functional communication unit has been in the spotlight of neuroscientific inquiry (Cowan et al., 2001). In fact, many studies have demonstrated that synaptic events, such as changes in molecular composition, structure, efficacy, and potentiation, play important roles in brain functions including memory formation, perception, and other complex behaviors (Tsien et al., 1996; Markram et al., 1997; Malinow and Malenka, 2002; Russo et al., 2010; Caroni et al., 2014). An important focus has been to visualize the synapse and to measure its activity. In 1954, DeRobertis and Palay first observed synapses independently by electron microscopy (EM) and George Gray suggested there may be different types of synapses, i.e., excitatory and inhibitory (De Robertis and Bennett, 1955; Palay and Palade, 1955; Palay, 1956; Gray, 1959). EM provides enough resolution for nanometer-scale imaging of the synaptic ultrastructure, something that cannot be achieved by light microscopy (LM) because of its diffraction limit. However, despite recent advances that have reduced the time needed for image acquisition and reconstruction, EM remains inherently time-consuming, laborintensive, and volume-limited for large neural circuits.

With the recent engineering of fluorescent proteins (FPs) and new developments in light-favoring tissue clearing and advanced optical methods, LM is rising as a potent alternative tool for investigating individual synapses in the context of neural networks (Gray, 1959; Keller et al., 2008; Kim et al., 2012; Tomer et al., 2012; Chung et al., 2013; Richardson and Lichtman, 2015). Imaging the synapse with LM by using newly engineered FPs tagged to synaptic proteins or targeted to synaptic structures enables fine-resolution illumination of synaptic anatomy and function in large neural circuits, possibly in real-time. In this review, we describe recent FP-based molecular tools for imaging individual synapses and synaptic connectivity in the contexts of single- and dual component synaptic detection. We also identify crucial technologies: gene delivery of molecular synapse detector; tissue clearing for whole-brain imaging; and computational analysis, whose parallel development has potential to bridge synaptic sensor engineering and systems neuroscience. We end by proposing a scheme of technological integration for synaptic neuroscience at the systems level.

### SINGLE COMPONENT SYNAPTIC DETECTION

The discovery of green fluorescent protein (GFP) and its derivatives revolutionized the visualization of biological phenomena, including the individual synapse and its functions. GFP and other FPs are relatively inert and small (27 kDa) and can be used to tag synaptic proteins while minimally interfering with their normal functions. In fact, the distribution, trafficking, and physiological changes of synaptic proteins caused by neural activity became evident in the last two decades, largely through observations of synaptic proteins, such as synaptophysin (Li and Murthy, 2001) vesicle-associated membrane protein 2 (VAMP2; Ahmari et al., 2000), postsynaptic density protein-95 (PSD-95; Nelson et al., 2013) calcium/calmodulin-dependent kinase II (CaMKII; Shen et al., 2000), α-amino-hydroxy-5 methyl-4-isoxazolepropionic acid receptor (AMPAR; Zamanillo et al., 1999) and so forth, that had been tagged with FPs. Recently, sophisticated molecular engineering has allowed even more precise and detailed visualization of synaptic structure, composition, and physiology (Chen et al., 2014; Fortin et al., 2014; **Figure 1**, **Table 1**).

### pH-Sensitive FPs for Visualizing Vesicle Release: pHluorin, pHTomato, and pHuji

Although previous, straightforward, FP-based detection of synaptic distribution revealed many important details of synaptic physiology, synaptic vesicle release/recycling and neural activitydriven changes in membrane-bound synaptic proteins are difficult to be detected by regular FP-tagging. Special sensors of vesicle secretion and neurotransmission have been developed by linking vesicle membrane proteins with pH-sensitive mutants of GFP called pHlourin (Miesenböck et al., 1998). The fluorescence intensity of pHluorin largely depends on the pH of its biochemical environment: in acidic environments with pH below 6.5, pHluorin is mostly nonfluorescent in 480 nm of light illumination, but becomes highly fluorescent in neutral environments with pH around 7.4. This special feature of pHluorin was achieved by several mutations on residues important for the proton-relay of tyrosine 66 in the chromophore (S147D, N149Q, T161I, S202F, and Q204T). These amino acid substitutions, which set the pKa of pHluorin to around 7.0, can facilitate the pH-dependent switching of the electrostatic environment of the chromophore, allowing the pHdependent changes in fluorescent intensity (Sankaranarayanan and Ryan, 2000). When pHluorin is fused to presynaptic vesicle proteins such as VAMP2 (Miesenböck et al., 1998), synaptophysin (Zhu et al., 2009), and vesicular glutamate transporter (vGluT; Voglmaier et al., 2006), the release and recycle of synaptic vesicles can be monitored as the pH inside the synaptic vesicles (∼5.5) transitions to the pH of the extracellular environment (∼7.4). Thus, by tracking changes in pHluorin fluorescence intensity, one can detect real-time presynaptic exocytosis in living, active neurons. Similarly, postsynaptic endo-/exo-cytosis and related receptor dynamics can be visualized by pHluorin-fused mGluRs, for instance (Pelkey et al., 2007).

More recently, red pH-sensitive FPs such as pHTomato (Li and Tsien, 2012) and pHuji (Shen et al., 2014) have been developed, allowing for the simultaneous monitoring of multiple synaptic activities when combined with the green GFP-based sensors (e.g., GCaMP). pHTomato (pKa ∼ 7.8) was derived from mStrawberry by introducing six mutations (F41T, F83L, S182K, I194K, V195T, and G196D). Synaptophysin-fused pHTomato (sypHTomato) has been shown to be suitable for simultaneously monitoring the fusion of synaptic vesicles and, when paired with GCaMP3, postsynaptic Ca2<sup>+</sup> changes in living neurons. In addition, multiple synaptic events have been successfully measured by using sophisticated combinations of these pHsensitive FP-fused synaptic proteins and spectrally distinct variants of optogentic stimulators (ChR2-T2A-vGluT-pHlourin

FIGURE 1 | Scheme of single component synapse detectors. (A) Graphical depiction of the conventional green fluorescent protein (GFP) tagging scheme and summary of the expressed carrier-sensor construct. GFP (green circle) and other fluorescent proteins (FPs) are directly tagged to synaptic proteins in the presynaptic terminal or postsynaptic spines. Major targets for tagging include presynaptic vesicle proteins (e.g., vesicle-associated membrane protein 2 (VAMP2) and synaptophysin), postsynaptic receptors (e.g., α-amino-hydroxy-5-methyl-4-isoxazolepropionic acid receptor, AMPAR) and postsynaptic density protein-95 (PSD-95). (B) pH-sensitive FP mutants are fused to synaptic vesicle membrane proteins, such as VAMP2, to visualize vesicle secretion/recycling and neurotransmission. A pH-sensitive GFP variant, pHluorin does not fluoresce (gray circle) when inside the acidic chemical environment of the synaptic vesicle, but becomes highly fluorescent (green circle) when the vesicle is released and exposed to the neutral extracellular environment. (C) Inhibition of Synaptic Release with CALI (InSynC): attached to target SNARE proteins, molecular actuators such as mini small singlet oxygen generator (miniSOG; light blue filled-in circle) selectively inactivate specific synaptic proteins that regulate vesicle release and other synaptic events. When illuminated with blue light, miniSOG stimulates generation of reactive oxygen species (small, dark blue filled-in circles), which then oxidizes susceptible amino acid residues in target vesicle proteins and deactivates protein functions. (D) TimeSTAMP effectively tracks spatiotemporally controlled protein synthesis and trafficking in living neurons. In the presence of a membrane-permeable protease inhibitor, NS3 protease (gray oval) activity is inhibited, and Venus C-terminus (Venus CT) and Venus N-terminus (Venus NT) reconstitute as fluorescent Venus (yellow circle). Reconstituted Venus accumulates in the postsynaptic spine, the trafficking destination of the fused PSD-95. When the protease inhibitor is present, however, NS3 protease cleaves the protease target sites (gray circles flanking NS3), preventing Venus reconstitution.

and VChR1-T2A-synpHTomato). However, the pH-sensitivity of pHTomato is relatively low (3-fold change in pH 5.5–7.5). Thus, a new red pH-sensitive FP, named pHuji, has been more recently derived from mApple (Shaner et al., 2008) by including a K163Y mutation, resulting in high pH sensitivity (20-fold change in pH 5.5–7.5). The use of different colored pH-sensitive FPs together with a Ca2<sup>+</sup> indicator and/or a spectrally distinct optogenetic modulator offers a new promising readout system for complex, coordinated synaptic events.

## Inhibition of Synaptic Release with CALI (InSynC)

Beyond merely visualizing the distributions and endo-/exocytosis of synaptic proteins by tagging them with FPs and their pHsensitive variants, genetically encoded chromophore-assisted light inactivation (CALI) has been developed to selectively inactivate specific synaptic protein functions that regulate synaptic events such as synaptic release (Lin et al., 2013). CALI is based on light-induced generation of reactive oxygen and the consequent inactivation of nearby attached synaptic proteins. Its original agents were synthetic chromophores for example malachite green (Jay, 1988), fluorescein (Beck et al., 2002), FlAsH (Marek and Davis, 2002), ReAsH (Tour et al., 2003), and eosin (Takemoto et al., 2011) and FPs such as KillerRed (Bulina et al., 2006). To precisely inhibit synaptic release with improved target specificity and inactivation efficiency compared to these CALI agents, inhibition of synaptic release with CALI (InSynC) has been recently developed using a newly engineered flavoprotein called mini small singlet oxygen generator (miniSOG), fused with the SNARE proteins VAMP2 and synaptophysin (Lin et al., 2013). miniSOG was originally derived from the light, oxygen, and voltage 2 (LOV2) domain of phototropin, a blue light photoreceptor (Shu et al., 2011). Under blue light illumination, this photoreceptor binds to and excites flavin mononucleotide (FMN), which then functions as an oxygen generator in cells. miniSOG contains the single amino acid substitution of FMN-binding residue Cys426 to Gly on the LOV2 domain, allowing for more efficient energy transfer to FMN, and contains further mutations for enhanced brightness. When illuminated by blue light, synaptic proteins can be selectively inactivated by miniSOG-mediated oxidization of susceptible residues such as tryptophan, tyrosine, histidine, cysteine, and methionine. InSynC by miniSOG can selectively inhibit vesicle release at individual synapses in vitro and in vivo thanks to the high efficiency of its light-induced oxygen generation, independence of exogenous cofactors, and small size (106 residues, 14 kDa; Lin et al., 2013). However, further engineering is required for investigating the functional dynamics of synaptic circuits with physiologically relevant temporal resolutions, as inactivation by InSynC persists relatively long (∼1 h) after light stimulation.

### Time-Specific Tagging for the Age Measurement of Proteins (TimeSTAMP)

Another new strategy beyond merely visualizing synaptic proteins by tagging with FPs is time-specific tagging for the age measurement of proteins (TimeSTAMP; Lin et al., 2008). As spatiotemporally controlled protein synthesis and trafficking are critical for synaptogenesis, synaptic connectivity, and long-lasting changes in synapses, TimeSTAMP is beneficial for tracking important, newly synthesized synaptic proteins such as PSD-95 and CaMKII in living neurons. TimeSTAMP is a drug-controllable, time-specific tagging strategy based on the hepatitis C virus NS3 protease and its cell-permeable inhibitor BILN-2061. PSD-95-GFP, for example, was fused to NS3 protease flanked by NS4A/B target sites and the C-terminal HA tag (PSD-95-GFP-TS-HA). In the absence of the specific inhibitor BILN-2061, NS3 protease cleaves the NS4A/B target sites allowing the C-terminal HA-tag to be cleaved from PSD-95-GFP and degraded; but drug application will allow the HA-tag to be accumulated. This allows distinguishing between newly and previously synthesized PSD-95 by measuring HA/GFP signals at a time defined by the drug application. For the TimeSTAMP strategy, the NS3 protease domain was chosen because it is small (19 kDa) and monomeric, specific to its substrate, and not cytotoxic, and most of all, NS3 shows high selectivity and efficiency of its cell-permeable inhibitor.

Although the first version of TimeSTAMP has worked successfully in primary hippocampal neurons to track PSD-95 accumulation during synaptic growth, and in Drosophila for whole-brain mapping of newly synthesized CaMKII, it required post hoc immunostaining against epitope tags, limiting the benefits of pulse-chase labeling through time. Thus, TimeSTAMP2 has been introduced, replacing HAtag with split-fluorescence proteins (e.g., Yellow fluorescent protein, YFP; Orange fluorescent protein, OFP). In fluorescent TimeSTAMP2, for instance, Venus yellow fluorescent protein is separated by an NS3 domain flanked by NS4A/B target sites into VenusNT (1-158 aa) and VenusCT (159-238 aa) for drug-dependent fluorescence. In the absence of BILN-2061, VenusCT is cleaved and degraded by the active NS domain resulting in no yellow fluorescence, while VenusNT and CT are reconstituted as fluorescent forms after drug application. This allows optical pulse labeling of synaptic proteins such as PSD-95 and Neuroligin with a drug-defined temporal resolution. Additionally, photo-oxidizing TimeSTAMP using miniSOG inserted into TS:YFP can be used to visualize new proteins at an EM-based ultrastructural level (Butko et al., 2012).

### DUAL COMPONENT SYNAPTIC DETECTION

Thus far, we have described methods for detecting synapses by labeling single components, either pre- or post-synaptic. Although these single-component tools allow for detecting, measuring, and manipulating synaptic structures and activities, they do not address the critical fact that synapses are bilateral microstructures involving both presynaptic terminal and postsynaptic density. For reliable synapse detection, therefore, several recent studies introduced new methods for labeling synaptic interactions between pre- and post-synaptic components, such as using split-FPs or enzymes to label neurexin-neuroligin interactions. Here we review dual-component methods using handshaking-like transmembrane molecular interaction across the synaptic cleft, with particular attention to GFP Reconstitution Across Synaptic Partners (GRASP; **Figure 2**, **Table 1**).

### Mammalian GFP Reconstitution Across Synaptic Partners (mGRASP)

To detect particular synaptic connections with LM, the Mammalian GFP Reconstitution Across Synaptic Partners (mGRASP) technique takes advantage of the complementarity of two non-fluorescent split-GFP fragments, each of which is tethered specifically to the pre- and postsynaptic membrane, respectively (Kim et al., 2012). When two neurons, each expressing one of the fragments, are closely opposed across a

synaptic cleft, the split fragments are reconstituted as fluorescent GFP in that location. This dual component synapse detection bypasses the Abbe's diffraction limit and allows relatively rapid and accurate synapse mapping with LM. This method takes advantage of the fast folding kinetics and stability after maturation of superfolder GFP (Pédelacq et al., 2006), split into two fragments, namely spGFP1-10 (first 214 residues comprising ten β barrels) and spGFP11 (16 residues, 11th βbarrel strand). The GRASP technique was initially implemented in C. elegans (Feinberg et al., 2008; Park et al., 2011) and later in Drosophila (Gordon and Scott, 2009; Fan et al., 2013; Gorostiza et al., 2014). In the original GRASP constructs, membrane carriers (human CD4) for split-GFP fragments were not synapse-specific—fluorescence could arise wherever any membranes expressing fragment pairs were closely apposed. This became a critical issue for mammalian synapse detection because the mammalian nervous system contains much more compactly intermingled neurites than the invertebrate nervous system.

In a previous study, we engineered and successfully applied mGRASP for mapping fine-scale synaptic connectivity in the mouse brain (Kim et al., 2012). The primary features of mGRASP include targeting specific synapses, and matching the approximately 20 nm-wide synaptic cleft without gross changes in endogenous synaptic organization and physiology. Based on published sequences from NCBI, we generated synapsespecific chimeric carriers. Both the pre- and postsynaptic mGRASP components share a common six-subcomponent framework: (1) a signal peptide; (2) a split-GFP fragment; (3) an extracellular domain; (4) a transmembrane domain; (5) an intracellular domain; and (6) a fluorescent protein for neurite and soma visualization. The presynaptic mGRASP component consists of the signal peptide of nematode βintegrin (PAT-3, residues 1–29), GFP β-strand 11 (spGFP11, 16 residues), two flexible GGGGS linkers, the extracellular and transmembrane domains of human CD4-2 (residues 25–242). Rat neurexin-1β (residues 414–468) containing the PDZ-binding motif constitutes the intracellular domain for maintenance of correct localization at the presynaptic site. mCerulean is fused to the intracellular end of the presynaptic mGRASP to visualize axonal expression. The postsynaptic mGRASP component is based primarily on mouse neuroligin-1, which interacts with presynaptic adhesion proteins including β-neurexins, and mediates the formation and maintenance of synapses between neurons. To prevent nonspecific synaptogenesis through interactions with its


endogenous partner, neurexin, the extracellular esterase domain of neuroligin is deleted in the main skeleton of the postsynaptic mGRASP. Similar to presynaptic mGRASP, post-mGRASP is composed a signal peptide from the esterase-truncated neuroligin-1 (residues 1–49), GFP βstrand 1-10 (spGFP1-10, 648 residues), the extracellular, transmembrane, and intracellular regions of neuroligin (71, 19, and 127 residues, respectively), followed by the selfcleavable 2A peptide-fused dTomato for visualizing postsynaptic neuronal morphology. This optimized mGRASP enabled the comprehensive synaptic connectivity mapping of hippocampal CA3-CA1 and identified spatially structured patterns of synaptic connectivity (Druckmann et al., 2014). It is important to note that brain-wide synapse detection for comprehensive fine-scale mapping becomes achievable not only by advanced molecular engineering to label the synapse such as mGRASP, but also by appropriately engineered computational analysis (Feng et al., 2012, 2015).

Current GRASP technology has proved to be a tool suitable for rapid and accurate mapping of synaptic connectivity in nematode (Feinberg et al., 2008), fruit fly (Gordon and Scott, 2009), and mouse (Kim et al., 2012; Druckmann et al., 2014). Yet, further improvements to GRASP such as multi-colored FPs for analyzing convergent synaptic inputs, various carriers for neural activities, and tailored computational analyses will provide a clear overview of complex synaptic connectivity and its operation. We next discuss recent efforts in these directions.

### Engineering FPs for Multi-Color GRASP

A neuron oftentimes receives multiple inputs from different presynaptic neurons of distinct cell types and/or various brain areas. For comprehensive mapping of complex synaptic circuits, multiple synaptic detection is beneficial, as described earlier in the section describing pH-sensitive FPs. The current version of GRASP restricts convergent synaptic mapping because it relies on only a single pair of spGFP fragments such that spectral overlap of its signals hinders simultaneous imaging with previously well-established GFPbased tools such as GCaMP. Therefore, the multi-color GRASP (XRASP) components have been developed recently (Macpherson et al., 2015; Li et al., 2016). Given that Cyan fluorescent protein (CFP), GFP, and YFP have identical structures except for several residues in the chromophore located in beta barrels of 1-10, C-RASP and Y-RASP were generated by color-shifting mutations in spGFP1-10 (Y66W, S72A, F145A<sup>∗</sup> , N146I, and H148D for C-RASP, <sup>∗</sup> additional mutation in Li et al., 2016, T65G, V68L, S72A and T203Y for Y-RASP) paired with the unaltered spGFP11 (Li et al., 2016). Multi-color GRASP (XRASP) was tested in vivo in several circuits, including the Kenyon cells of the mushroom body, and projection neurons of the thermosensory and olfactory systems in the fruit fly. Reconstructed CFP and YFP signals showed minimal spectral overlap with the GRASP emission and excitation spectra. Extended choices of multicolor GRASP (XRASP) will allow simultaneous imaging of multiple, convergent connectivity and functional activity. However, more red-shifted XRASP is required for in vivo 2-photon imaging together with Ca2<sup>+</sup> indicators and for fine-scale synapse labeling from multiple inputs within the single neuron, because it has proved difficult to separate CFP/GFP/YFP signals in the complex mammalian nerve system.

### Engineering Carriers for Activity-Dependent GRASP

To understand functional organizations underlying complex brain functions that go beyond structural connectivity patterns, it will be essential to identify active synaptic connectivity at defined times and conditions. For the activity-dependent GRASP system, synaptobrevin (syb), a key constituent of synaptic vesicle membrane, was used as a synaptic carrier instead of constantly membrane-targeted carriers in the previous GRASP systems. The straightforward fusion of syb and spGFP1-10 (syb:spGFP1-10) with the original CD4:spGFP11 can together form activity-dependent GRASP, called the syb:GRASP system. Given activity-dependent interactions of syb with SNARE-SM protein complex triggering synaptic vesicle release, syb:GRASP showed preferential labeling of active synapses as a boost of GRASP fluorescence signals in well-studied thermosensory and olfactory circuits in the fruit fly (Macpherson et al., 2015). This new strategy for mapping active synaptic connectivity might expedite the mapping of functional connectivity at the synapse level in a way previously achieved only by difficult combinations of Ca2<sup>+</sup> imaging and exhaustive EM reconstruction (Bock et al., 2011; Briggman et al., 2011). Yet, these new techniques raise basic concerns about possible side effects caused by the overexpression of key synaptic vesicle proteins, and need further optimization before they can be applied to mammalian networks.

### FP- and Enzyme-Based Visualization of Neuroligin-Neurexin Interaction

An alternative way to detect synapses has been suggested by imaging neurexin-neuroligin interactions. Presynaptic neurexin and postsynaptic neurolign are transmembrane adhesion proteins and their interaction at the synaptic cleft has been believed to be a key process for synaptic formation, maintenance, and connectivity (Li and Sheng, 2003; Graf et al., 2004; Chen et al., 2010; Krueger et al., 2012). Therefore, it is thought that identifying sites of neurexin-neuroligin interactions could provide selective visualization of synapses. Similar to the GRASP approaches, this method, based on neuroligin-neurexin interactions using splitGFP, is called SynView. It is composed of spGFP1- 10-inserted neurexin-1β between residues N275-D276 or A200-G201 for the presynaptic component, and the split-GFP11-inserted neuroligin-1 in the C-terminal end of the extracellular esterase (between Q641-Y642) for the postsynaptic component (Tsetsenis et al., 2014). Another method for visualizing neuroligin-neurexin interactions is based on mutated bacterial lipoic acid ligase (LpIA) and lipoic acid acceptor peptide (LAP) tags (Uttamapinant et al., 2010, 2012), and is called Interaction-Dependent Probe Incorporation Mediated by Enzymes (ID-PRIME; Liu et al., 2013). Using the same principle of SynView, ID-PRIME was designed to detect trans-synaptic contacts of neurexin and neuroligin enzymatically using lipoic acid. These two methods successfully imaged the direct interactions of neurexin-neuroligin synaptic adhesive molecules. However, these methods seem restricted to particular investigations of the dynamics of neurexinneuroligin, rather general synaptic mapping. In addition, it is known that overexpressing these synaptic adhesion molecules causes substantial structural and physiological perturbations of normal synaptic compartments and cleft structures.

### BRIDGING TOOLS BETWEEN MOLECULAR SYNAPSE DETECTORS AND BRAIN-WIDE SYNAPSE MAPPING

Thus far, we have described methods for detecting synapses focusing on molecular engineering. To exert these genetically encoded synapse detectors to brain-wide synaptic mapping at the system level, there need to be critically partnered technologies such as gene delivery, advanced imaging, and digital representation platform that are appropriate for neural system (**Figure 3**).

### Gene Delivery System for Synapse Detectors in the Mouse Brain

Improving the targeting specificity for types of cells and scaling up the scope of synaptic sensor delivery are among the crucial initial steps needed to expand single-synapse level analysis to systems level neuroscience. Gene delivery technologies, particularly for the nervous system, have developed at a breathtaking pace over the past decades, establishing methodologies that can be classified into two main categories: germ line manipulation and viral injection.

### Genetic Manipulation-Based Gene Delivery

To avoid side effects that can sometimes be caused by the overexpression of FP-tagged synaptic proteins, and to mimic expression patterns of endogenous proteins, gene knockin (KI) technology has been used to substitute wild-type genes with FP-tagged copies. For synaptic detectors based on universal synaptic proteins such as PSD-95, however, this standard KI method leads to expression of FP-tagged synapse detectors globally, throughout the brain, which makes it difficult to acquire high resolution images of a particular cell type. Therefore, a recent study introduced a conditional KI strategy called endogenous labeling via exon duplication (ENABLED; Fortin et al., 2014). In the ENABLED strategy, a knocked-in gene cassette is composed of the floxed last exon and 3'UTR of PSD-95 followed by its mVenus-tagged duplicated last exon and 3'UTR. The FP-tagged duplicated gene will be selectively expressed only in Cre-expressing neurons because its translation is designed to be blocked by translation stop signals in the endogenous copy in the absence of Cre recombinase. When this PSD-95-ENABLED mouse line is crossed with a dopaminergic cell-type specific DAT-Cre line, for instance, PSD-95mVenus is expressed specifically in dopaminergic neurons. PSD-95-ENABLED was shown to functionally replace wild-type PSD-95 and to sparsely label a particular cell-type, allowing insights into the detailed distribution and dynamics of synapses. In applying this strategy to a wide variety of synaptic proteins, one concern with using the ENABLED strategy is that it is limited to synaptic proteins that are compatible with C-terminal FPtagging.

Thanks to incredibly fast developments and improvements in genetic manipulation technologies such as clustered regularlyinterspaced short palindromic repeats (CRISPR) and effector nucleases Cas system (Cong et al., 2013; Wang et al., 2013; Fujii et al., 2014; Aida et al., 2015), the generation of synapse detector KI mouse lines is becoming time- and cost-efficient, and is dramatically facilitating synapse mapping.

### Viral System-Based Gene Delivery

Spatially targeted gene delivery into the mature brain can be achieved through stereotactic microinjection of viral vectors expressing FP-tagged proteins. Diverse virus families have been recruited into this effort: Retroviridae (e.g., lentivirus), Parvoviridae (e.g., rAAV), Adenoviridae (e.g., canine adenovirus), and Alphaviridae (e.g., sindbis virus), including some that cross the synapse, such as Rhabdoviridae (e.g., rabies virus) and Herpesviridae (e.g., HSV-1 and pseudorabies virus; Nassi et al., 2015). Because of their low cytotoxicity and stable expression, lentiviruses and rAAVs are the most widely used viral vectors in neuroanatomical tracing studies and human clinical trials testing gene therapy. In fact, spatially restricted injection of Cre-(in)dependent rAAV vectors have been used for mGRASP-assisted synaptic mapping.

In parallel, ongoing efforts include searching for and engineering new types of virus to allow transduction efficiency/ specificity and retrograde infection. Canine adenovirus has drawn attention because of its strong retrograde transport capability and relatively large payload size (∼30 kb); further, several successful applications of Canine adenovirus in the mouse brain suggest a powerful complement to the lentivirus and rAAV (Bru et al., 2010; Ekstrand et al., 2014; Junyent and Kremer, 2015; Schwarz et al., 2015). Additionally, systemic delivery of viral vectors in animal models has proved to be effective and safe for various serotypes of AAV (Foust et al., 2009; Bevan et al., 2011; Gray et al., 2011; Yang et al., 2014; Deverman et al., 2016).

### Advanced Brain-Wide Imaging and Digital Representation of Synaptic Detectors

Once FP-based synaptic detectors are introduced into the brain as described above, appropriate brain-wide imaging and digital representation technologies are necessary to map synaptic connectivity. Happily, there has been remarkable progress in tissue clearing methodology such as BABB (Dodt et al.,

GRASP (mGRASP) constructs and newly engineered serotypes that have more selective tropism and transduction efficiency (e.g., AAV-PHP.B). Local tissue or systemic injection of such viral systems can lead to flexible, versatile gene delivery in mature organisms. (B) Combination of light-favoring brain clearing, whole-brain imaging, and computational techniques for three-dimensional synapse mapping enables single-synapse level analysis of synaptic profiles across the whole brain. Further improvements in lipid extraction, refractive index matching, advanced light-sheet microscopy, and large-scale data processing and 3D reference space generation will accelerate systems neuroscience at the synaptic scale.

2007), CUBIC (Susaki et al., 2014) 3DISCO (Ertürk et al., 2012), immunolabeling-enabled three-dimensional imaging of solvent-cleared organs (iDISCO; Renier et al., 2014), SeeDB (Ke et al., 2013) and CLARITY (Chung et al., 2013; Tomer et al., 2014) that are needed to prepare the intact brain for imaging while avoiding distortions caused by physical sectioning. These new clearing techniques, together with advanced optical methods such as fluorescent selective plane illumination microscopy (fSPIM; Huisken et al., 2004), will allow high-throughput whole brain imaging. Also, once images are acquired, digital representation programs are necessary to reliably extract synaptic wiring information from the images, and to bring data from different sections and animals into register with one another (Johnson et al., 2010; Oh et al., 2014). Improvements in this software will be greatly beneficial.

In our view, new clearing methods, optics, and tailored computational analysis platforms together with advanced synaptic detectors such as mGRASP are very promising developments for the complete mapping of mammalian synaptic connectivity.

### CONCLUSION AND PERSPECTIVE

Here we reviewed new FP-based synapse detection techniques, which are useful for rapidly imaging individual synapses and synaptic connectivity in the whole brain with LM. Recent technical developments have allowed a focus of neuroscience to move from individual synapses to ensemble interactions among neurons of various cell types through multiple synaptic pathways. Comprehensively mapping individual synapses in the whole

### REFERENCES


brain will provide firm grounds for further anatomical and functional studies and such mapping is essential for analyzing large-scale information processing phenomena. We believe that rapid, scalable synaptic cartography with the triad of singlesynapse resolution, brain-wide scope, and cell-type-specific connectivity requires a synergistic combination of advanced technologies, and that FP-based neurosensors are the key to this grand integrative project.

### AUTHOR CONTRIBUTIONS

HL, WCO, JS, and JK wrote this manuscript.

### ACKNOWLEDGMENTS

HL, WCO, JK, and JS are supported by the Korea Institute of Science and Technology (KIST) Institutional Program (2E26190 and 2N41660, respectively) and the Samsung Science and Technology Foundation.


structural plasticity in nucleus accumbens. Trends Neurosci. 33, 267–276. doi: 10.1016/j.tins.2010.02.002


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Lee, Oh, Seong and Kim. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Organelle-Specific Sensors for Monitoring Ca2+ Dynamics in Neurons

### Seok-Kyu Kwon, Yusuke Hirabayashi and Franck Polleux\*

Department of Neuroscience, Mortimer B. Zuckerman Mind Brain Behavior Institute, Kavli Institute for Brain Science, Columbia University Medical Center, New York, NY, USA

Calcium (Ca<sup>2</sup><sup>+</sup>) plays innumerable critical functions in neurons ranging from regulation of neurotransmitter release and synaptic plasticity to activity-dependent transcription. Therefore, more than any other cell types, neurons are critically dependent on spatially and temporally controlled Ca<sup>2</sup><sup>+</sup> dynamics. This is achieved through an exquisite level of compartmentalization of Ca<sup>2</sup><sup>+</sup> storage and release from various organelles. The function of these organelles in the regulation of Ca<sup>2</sup><sup>+</sup> dynamics has been studied for decades using electrophysiological and optical methods combined with pharmacological and genetic alterations. Mitochondria and the endoplasmic reticulum (ER) are among the organelles playing the most critical roles in Ca<sup>2</sup><sup>+</sup> dynamics in neurons. At presynaptic boutons, Ca<sup>2</sup><sup>+</sup> triggers neurotransmitter release and synaptic plasticity, and postsynaptically, Ca<sup>2</sup><sup>+</sup> mobilization mediates long-term synaptic plasticity. To explore Ca<sup>2</sup><sup>+</sup> dynamics in live cells and intact animals, various synthetic and genetically encoded fluorescent Ca<sup>2</sup><sup>+</sup> sensors were developed, and recently, many groups actively increased the sensitivity and diversity of genetically encoded Ca<sup>2</sup><sup>+</sup> indicators (GECIs). Following conjugation with various signal peptides, these improved GECIs can be targeted to specific subcellular compartments, allowing monitoring of organelle-specific Ca<sup>2</sup><sup>+</sup> dynamics. Here, we review recent findings unraveling novel roles for mitochondria- and ER-dependent Ca<sup>2</sup><sup>+</sup> dynamics in neurons and at synapses.

#### Edited by:

Marc Fivaz, Duke NUS Graduate Medical School, Singapore

Reviewed by:

Lucas Pozzo-Miller, University of Alabama at Birmingham, USA William N. Green, University of Chicago, USA

> \*Correspondence: Franck Polleux fp2304@cumc.columbia.edu

Received: 13 June 2016 Accepted: 30 August 2016 Published: 16 September 2016

#### Citation:

Kwon S-K, Hirabayashi Y and Polleux F (2016) Organelle-Specific Sensors for Monitoring Ca2+ Dynamics in Neurons. Front. Synaptic Neurosci. 8:29. doi: 10.3389/fnsyn.2016.00029 Keywords: synapse, mitochondria, endoplasmic reticulum, circuit function, calcium dynamics

### INTRODUCTION

Calcium (Ca2+) ions govern prevalent physiological processes in various cell types (Rizzuto and Pozzan, 2006; Clapham, 2007). This is especially prominent in excitable cells like neurons where Ca2<sup>+</sup> influx through the plasma membrane and release of Ca2<sup>+</sup> from internal stores transduce the effects of changes in membrane polarization and therefore mediate faithful transfer or storage of information over various timescales (milliseconds to minutes/hours). Therefore, regulation of intracellular Ca2<sup>+</sup> homeostasis is central to the proper function of neuronal circuits. The maintenance of baseline levels of intracellular Ca2<sup>+</sup> levels is regulated in part through exchangers and pumps such as the plasma membrane Ca2+-ATPase (PMCA pump), the Na+/Ca2<sup>+</sup> exchanger (NCX), and the Na+/Ca2+-K<sup>+</sup> exchanger (NCKX) which extrude Ca2<sup>+</sup> through the plasma membrane into the extracellular space. In addition to these mechanisms, intracellular organelles, such as mitochondria and endoplasmic reticulum (ER), are able to regulate cytoplasmic Ca2<sup>+</sup> ([Ca2+]c) through mitochondrial calcium uniporter (MCU) and smooth endoplasmic reticulum Ca2+-ATPase (SERCA), respectively.

In neurons, mitochondria and ER play important physiological roles via [Ca2+]<sup>c</sup> regulation, thereby diverse synaptic functions including basal synaptic transmission, presynaptic short-term plasticity, and long-term plasticity can be regulated by these organelles (Verkhratsky, 2005; Bardo et al., 2006; Mattson et al., 2008; Vos et al., 2010). In addition, impaired Ca2<sup>+</sup> homeostasis in the nervous system has been proposed to play an important function in the physio-pathological mechanisms underlying Alzheimer's disease, Parkinson's disease, and spinocerebellar ataxia (Verkhratsky, 2005; Mattson et al., 2008; Schon and Przedborski, 2011).

To monitor Ca2<sup>+</sup> dynamics, various fluorescent Ca2<sup>+</sup> dyes and genetically encoded Ca2<sup>+</sup> indicators (GECIs) were developed and applied both in vitro and in vivo. Also, GECIs tagged with target peptide sequences have allowed imaging of Ca2<sup>+</sup> dynamics in specific organelles (Rizzuto et al., 1992; Palmer et al., 2004; Palmer and Tsien, 2006).

Previously published reviews have already summarized the usefulness and limitations of various Ca2<sup>+</sup> sensors and GECIs applied to neuronal and non-neuronal cells (Palmer and Tsien, 2006; Knopfel, 2012; Tian et al., 2012; Rose et al., 2014). In this review, we only describe recently uncovered insights about Ca2<sup>+</sup> dynamics and its regulation by mitochondria and ER, and we discuss how these organelle-specific Ca2<sup>+</sup> sensors have been used for the exploration of the role of these subcellular compartments in the regulation of Ca2<sup>+</sup> homeostasis and synaptic function in neurons.

### UNVEILED SYNAPTIC FUNCTIONS OF MITOCHONDRIA-DEPENDENT Ca2+ HOMEOSTASIS

Mitochondrial Ca2<sup>+</sup> uptake has been studied since the 1950s from studies of rat heart muscle and kidney (Slater and Cleland, 1953; Deluca and Engstrom, 1961). In the nervous system, mitochondria were described at presynaptic terminals and dendrites of various neuronal subtypes using the light and electron microscope (EM) several decades ago (Bartelmez and Hoerr, 1933; Palay, 1956; Gray, 1963; Shepherd and Harris, 1998; Rowland et al., 2000). In axons, mitochondria are short and sparsely distributed, and interestingly, several studies showed that half of presynaptic boutons are occupied by mitochondria (Shepherd and Harris, 1998; Kang et al., 2008). In contrast, dendritic mitochondria have tubular shapes and they are rarely observed in postsynaptic spines in the excitatory neurons (Sheng and Hoogenraad, 2007; Kasthuri et al., 2015).

At presynaptic boutons and terminals, synaptic vesicle (SV) fusion with the plasma membrane occurs following increase of [Ca2+]<sup>c</sup> following opening of voltage-sensitive Ca2<sup>+</sup> channels (VSCC) followed by Ca2<sup>+</sup> binding to sensors like synaptotagmins (Schneggenburger and Neher, 2005; Neher and Sakaba, 2008; Jahn and Fasshauer, 2012; Südhof, 2012). The ability of mitochondria to import Ca2<sup>+</sup> into the mitochondrial matrix ([Ca2+]m) plays a role in regulating presynaptic [Ca2+]c. This has been characterized in various species, neuronal cell types and circuits (**Figure 1A**). At the Drosophila neuromuscular junction (NMJ), the GTPase dMiro mutant lacks presynaptic mitochondria through impaired axonal transport (Guo et al., 2005; Wang and Schwarz, 2009). During prolonged stimulation, these mutants lacking presynaptic mitochondria displayed subtle, but significantly increased presynaptic Ca2<sup>+</sup> accumulation and display decrease forms of sustained synaptic transmission or synaptic ''fatigue'' (Guo et al., 2005). Drosophila Drp1 mutants also deplete presynaptic mitochondria at NMJ and exhibit elevated presynaptic Ca2<sup>+</sup> levels in resting and evoked states. However, spontaneous release (mini Excitatory junctional potential, mEJP) was not altered, but the evoked synaptic transmission was impaired during high frequency stimulation, and this defect was partially rescued by ATP (Verstreken et al., 2005) suggesting that mitochondria plays a role in synaptic transmission through their ability to generate ATP through oxidative phosphorylation. Although mitochondrial Ca2<sup>+</sup> uptake has limited effects on Drosophila NMJ neurons, in mammalian NMJ terminals, acute inhibition of mitochondrial Ca2<sup>+</sup> uptake causes rapid depression of the endplate potential (EPP) and increased asynchronous release (David and Barrett, 2003). Furthermore, in synapses of the mammalian central nervous system (CNS), mitochondria-dependent Ca2<sup>+</sup> uptake accelerates the recovery from synaptic depression in the calyx of Held (Billups and Forsythe, 2002). Other studies in mammalian hippocampal neurons claimed that impaired mitochondrial anchoring at presynaptic sites increases presynaptic Ca2<sup>+</sup> during repetitive stimulation and produces short-term facilitation (STF), and insulin-like growth factor-1 receptor (IGF-1R) signaling regulates resting mitochondrial Ca2<sup>+</sup> level and spontaneous transmission (Kang et al., 2008; Gazit et al., 2016). Although most pharmacological studies employed uncoupling agents as mitochondrial Ca2<sup>+</sup> influx blocker, which may affect ATP production, these reports support presynaptic control via mitochondrial Ca2<sup>+</sup> import (Ly and Verstreken, 2006). A recent study demonstrates that presynaptic boutons associated with mitochondria display lower levels of [Ca2+]<sup>c</sup> accumulation than presynaptic boutons not associated with mitochondria (Kwon et al., 2016). Furthermore, acute inhibition of mitochondria calcium import increased [Ca2+]<sup>c</sup> accumulation at presynaptic boutons occupied by mitochondria. In the same study, we demonstrate that this mitochondria-dependent regulation of [Ca2+]<sup>c</sup> plays an important role in regulating presynaptic release properties including spontaneous release, asynchronous release and short-term synaptic plasticity.

In addition to regulation of [Ca2+]<sup>c</sup> clearance, Ca2<sup>+</sup> release from mitochondria plays important roles at presynaptic sites (**Figure 1A**). Following the sustained high frequency stimulation, an enhancement of synaptic transmission lasting tens of seconds to minutes is observed and which is called post-tetanic potentiation (PTP; Zucker, 1989). Mitochondrial Ca2<sup>+</sup> release is suggested as one of the underlying mechanisms for this prolonged enhancement of synaptic transmission. Pharmacological inhibition of mitochondrial Ca2<sup>+</sup> uptake and release at the crayfish NMJ impaired PTP (Tang and Zucker, 1997; Zhong et al., 2001). Furthermore,

similar phenotypes were observed at mouse NMJ and hippocampal mossy fiber synapses with blocking the mitochondrial NCX, which mediates mitochondrial Ca2<sup>+</sup> release (García-Chacón et al., 2006; Lee et al., 2007).

In contrast to presynaptic boutons and terminals, the postsynaptic function of mitochondrial Ca2<sup>+</sup> regulation is less well-documented. In mouse hippocampal pyramidal neurons (Li et al., 2004), a minority (<5%) of dendritic spines contains mitochondria. Also, large branched spines in hippocampal CA3 contain mitochondria (Chicurel and Harris, 1992). However, a physiological role of these postsynaptic mitochondria is largely unknown. In general, mitochondria are distributed primarily in dendrite shaft and therefore localized microns away from the postsynaptic density, but might still be able to buffer [Ca2+]<sup>c</sup> mobilized through Ca2+-channels and glutamate receptors (Thayer and Miller, 1990; White and Reynolds, 1995; Wang and Thayer, 2002). This mitochondrial calcium import can stimulate tricarboxylic acid (TCA) cycle and might increase ATP production (Kann and Kovács, 2007) and may also regulate other ATP-dependent Ca2<sup>+</sup> pumps like PMCA and SERCA. While it is still unclear whether or not mitochondria play significant roles in regulating postsynaptic [Ca2+]<sup>c</sup> under physiological conditions of neurotransmission, they might play a role in pathophysiological contexts. For example, neurons lacking LRRK2, a protein associated with Parkinson's disease, show impaired dendritic Ca2<sup>+</sup> homeostasis through mitochondrial defects and thought to cause defective mitochondrial depolarization and reduction in dendritic complexity (**Figure 1B**; Cherra et al., 2013).

Overall, mitochondria-dependent Ca2<sup>+</sup> clearance and release in neurons plays important physiological and developmental roles pre- and post-synaptically but their functional importance seems to depend on the neuronal subtypes and the structure/size of the pre- and postsynaptic compartments.

### MITOCHONDRIAL Ca2+-IMAGING IN NEURONS AND AT SYNAPSES

To investigate organelle-specific Ca2<sup>+</sup> dynamics, various Ca2<sup>+</sup> sensors are developed (**Table 1**). One of the first method developed to monitor mitochondrial Ca2<sup>+</sup> dynamics was established using rhod-2, a cationic chemical Ca2+-binding fluorophore preferentially accumulating in the mitochondrial matrix presumably because of the highly negative membrane potential across the mitochondrial inner membrane (Minta et al., 1989). Then, in the calyx of Held, rhod-2 and rhod-FF (low affinity version) were used to visualize presynaptic mitochondrial Ca2<sup>+</sup> transient (Billups and Forsythe, 2002). However, these dyes cannot be precisely targeted to these organelles. Therefore, GECIs have recently become the preferred method to image Ca2<sup>+</sup> in specific organelles including mitochondria. For mitochondrial matrix localization, the targeting presequence of subunit VIII of human cytochrome c oxidase (COXVIII) was tagged to GECIs (Rizzuto et al., 1992). Mitochondria-targeted aequorin (mt-AEQ), a luminescent Ca2<sup>+</sup> indicator, was first employed to monitor the neuronal mitochondrial Ca2+, and this probe showed NMDA-induced mitochondrial Ca2<sup>+</sup> increase in hippocampal neurons (Baron et al., 2003). However, this probe needs a chemical reaction characterized by a modest turnover rate and has very limited dynamic range (Palmer and Tsien, 2006). Other GECIs have been developed and tested in various neuronal subtypes with the same targeting sequence. Mitochondrial-targeted ratiometric pericam (2mtRP) consists of circularly permutated Enhanced yellow fluorescent protein (cpEFYP) conjugated with Ca2+-responsive calmodulin (CaM) and its binding peptide (Nagai et al., 2001; Robert et al., 2001). This probe has a bimodal excitation spectrum and the relative emission intensity is dependent on Ca2+-binding. In hippocampal neurons, the use of 2mtRP described mitochondrial Ca2<sup>+</sup> uptake and also determined cytosolic Ca2<sup>+</sup> rise upon synaptic activation via dual imaging with cytosolic Ca2<sup>+</sup> dye (fura-red AM; Young et al., 2008). Other CaM conjugated cpEGFPs called GCaMPs (mito-GCaMP2, 2mtGCaMP6m, and mito-GCaMP5G) were used to monitor axonal mitochondrial Ca2<sup>+</sup> (Gazit et al., 2016; Kwon et al., 2016; Marland et al., 2016). Both sensors displayed action potential (AP)-dependent mitochondrial Ca2<sup>+</sup> import. In addition, red fluorescent GECIs by replacing cpEGFP with cpmApple or cpmRuby (mtRCaMP1e and LAR-GECO1.2) revealed mitochondrial Ca2<sup>+</sup> import simultaneously with cytosolic Ca2<sup>+</sup> (Akerboom et al., 2013; Wu et al., 2014).

However, these fluorescent proteins have some limitations, for example, they are affected by pH and mitochondrial matrix pH (pHm) can be changed by Ca2<sup>+</sup> influx (Abad et al., 2004; Poburko et al., 2011; Chouhan et al., 2012; Marland et al., 2016). In addition to this point, [Ca2+]<sup>m</sup> can span broad ranges (0.05–300 µM) depending on cell types and stimulation protocol (Arnaudeau et al., 2001; Palmer and Tsien, 2006). Thus, K<sup>d</sup> value for Ca2<sup>+</sup> of mitochondrial GECI should be considered for experimental purposes because high affinity (low Kd) sensors can be easily saturated by high [Ca2+]<sup>m</sup> and low affinity (high Kd) sensors may not be sensitive enough to detect small [Ca2+]<sup>m</sup> changes. Several studies reported low affinity mitochondrial Ca2<sup>+</sup> probes for avoiding saturation (Arnaudeau et al., 2001; Suzuki et al., 2014).

In conclusion, these mitochondria-targeted GECIs allow imaging of mitochondria Ca2<sup>+</sup> dynamics in neurons and have revealed interesting, synapse-specific properties of mitochondria in the regulation of [Ca2+]<sup>c</sup> and neurotransmitter release properties.

### REGULATION OF SYNAPTIC Ca2+ DYNAMICS BY THE ENDOPLASMIC RETICULUM

Neurons are among the most polarized cell types in our body and consists of a soma, relatively short dendrites and long axons. ER is found throughout the entire length of neuronal processes, and usually rough ER is prominent in the cell body and proximal dendrites, whereas smooth ER is dominant in distal dendrites, spines and axons (Spacek and Harris, 1997; Verkhratsky, 2005). ER imports and sequesters large amount of Ca2<sup>+</sup> ([Ca2+]er ∼500 µM) through SERCA and store-operated Ca2<sup>+</sup> entry (SOCE) mechanism (Verkhratsky, 2005; Bardo et al., 2006). Ca2<sup>+</sup> release from ER is mediated by two major mechanisms, called Ca2+-induced Ca2<sup>+</sup> release (CICR) and IP3-induced Ca2<sup>+</sup> release (IICR; Verkhratsky, 2005; Bardo et al., 2006). CICR is caused by the cytosolic Ca2<sup>+</sup> increase through N-Methyl-D-Aspartate receptors (NMDAR, GluN receptors) and voltage-gated Ca2<sup>+</sup> channels (VGCCs), whereas IICR is triggered by IP3, which is generated via activation of phospholipase C (PLC) depending on metabotropic glutamate receptors (mGluRs) or other receptors like receptor tyrosine kinases (**Figure 1**).

Ryanodine receptors (RyRs) are involved in CICR, and they have three major subtypes; RyR1, RyR2, and RyR3. All of these isoforms are detected in the brain, and show region-specific expression (Sharp et al., 1993; Furuichi et al., 1994; Giannini et al., 1995; Verkhratsky, 2005; Bardo et al., 2006; Baker et al., 2013). Similar to RyRs, IP<sup>3</sup> receptor (IP3R), which mediate IICR,


#### TABLE 1 | Organelle-specific Ca2+ sensors in neurobiology.

consist of three isoforms, IP3R1, IP3R2, and IP3R3, but IP3R1 is the dominant form in the brain (Sharp et al., 1993, 1999; Verkhratsky, 2005; Bardo et al., 2006; Baker et al., 2013).

Long-term synaptic plasticity is regulated by Ca2+ dependent signaling mechanisms such as Ca2+/calmodulindependent kinase II (CaMKII), calcineurin (a Ca2+-dependent phosphatase), protein phosphatase 1 (PP1) and protein kinase C (PKC; Malenka and Nicoll, 1999; Yang et al., 1999; Lüscher and Malenka, 2012). Therefore, Ca2<sup>+</sup> release from intracellular stores like the ER regulates long-term synaptic plasticity in specific circuits.

In cerebellar Purkinje cell dendrites, mGluR-IP3-dependent Ca2<sup>+</sup> increase is observed during parallel fiber (PF) stimulation and this mediates long-term depression (LTD) of PF-Purkinje cell pathway (Finch and Augustine, 1998; Takechi et al., 1998; Miyata et al., 2000; Wang et al., 2000). At synapses made by hippocampal Schaffer collateral (SC) onto CA1 pyramidal neurons, both long-term potentiation (LTP) and LTD are linked to IP3-dependent signaling (Oliet et al., 1997; Nishiyama et al., 2000; Raymond and Redman, 2002; Nagase et al., 2003). In addition, CICR is also observed in CA1 pyramidal neuronal spines, and LTD is abolished in RyR3-deficient mice and following application of RyR inhibitor (ryanodine) although the connection between CICR and LTP is controversial in SC-CA1 pathway (Reyes and Stanton, 1996; Emptage et al., 1999; Futatsugi et al., 1999; Sandler and Barbara, 1999; Kovalchuk et al., 2000; Nishiyama et al., 2000; Raymond and Redman, 2002). Hippocampal mossy fiber pathway (MF, dentate gyrus to CA3) shows IICR- and CICR-dependent LTP and LTD, however, there are conflicting results regarding the underlying mechanisms (**Figure 1B**; Yeckel et al., 1999; Itoh et al., 2001; Kapur et al., 2001; Mellor and Nicoll, 2001; Lauri et al., 2003; Lei et al., 2003).

Presynaptic ER-dependent Ca2<sup>+</sup> release is also detected and contributes to changes in neurotransmitter release properties and short-term synaptic plasticity at various inhibitory and excitatory synapses including basket cell to Purkinje cell synapses, hippocampal MF pathway, SC-CA1 and CA3-CA3 pyramidal neuron synapses (**Figure 1A**; Llano et al., 2000; Emptage et al., 2001; Liang et al., 2002; Galante and Marty, 2003; Lauri et al., 2003; Sharma and Vijayaraghavan, 2003; Unni et al., 2004; Mathew and Hablitz, 2008).

In addition to Ca2<sup>+</sup> efflux, Ca2<sup>+</sup> uptake by ER via SERCA pump affects STF at SC-CA1 presynapses and NMJ (**Figure 1A**; Castonguay and Robitaille, 2001; Scullin and Partridge, 2010; Scullin et al., 2010). Stromal interaction molecules (STIMs) and Orai1, which allow SOCE, are localized to neuronal compartment including dendritic spines, and impaired SOCE alters α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) trafficking, neuronal Ca2<sup>+</sup> signaling and LTP in CA1 pyramidal neuron and cerebellar Purkinje neurons (Baba et al., 2003; Hartmann et al., 2014; Korkotian et al., 2014; Garcia-Alvarez et al., 2015; Segal and Korkotian, 2015).

### IMAGING NEURONAL ER Ca2+ DYNAMICS

As mentioned above, ER contains high levels of Ca2+, therefore, in order to monitor Ca2<sup>+</sup> dynamics in the ER lumen, low affinity sensors were employed. Mag-Fura-2, a low affinity membranepermeable dye (K<sup>d</sup> = 53 µM), has been the first applied for neuronal ER Ca2<sup>+</sup> measurement, and after loading this dye in the cytoplasm and organelles, cytosolic dye was removed by perfusion with dye-free pipette solution (Solovyova et al., 2002). This allowed for the first time the visualization of caffeineinduced Ca2<sup>+</sup> release and reuptake in the ER of dorsal root ganglia (DRG) neurons (**Table 1**).

However, this method can non-specifically label internal compartments, therefore, genetically targeted sensors have been developed more recently for the visualization of ER-derived Ca2<sup>+</sup> dynamics (**Table 1**). The signal sequence of calreticulin, a Ca2+ binding protein in ER, and an ER retention sequence, KDEL, lead GECIs into the ER lumen, and to optimize for measuring a massive amount of [Ca2+]er, various mutations were applied to the CaM domain or EF-hand motif of existing GECIs for reducing their Ca2<sup>+</sup> affinity. A fluorescence resonance energy transfer (FRET)-based Ca2<sup>+</sup> sensor, D1ER, showed altered ER Ca2<sup>+</sup> leak function in hippocampal neurons of presenilin double knockout and Alzheimer's disease model mice (Zhang et al., 2010). Also, bioluminescence-based sensor, GFP-Aequorin protein (GAP), was modified and targeted to ER of DRG and hippocampal neurons, and showed 3- to 4-fold larger ratio change than D1ER (Rodriguez-Garcia et al., 2014). In addition, recently established GCaMP variants for ER Ca2<sup>+</sup> detection, calcium-measuring organelle-entrapped protein indicator one in the ER (CEPIA1er) and GCaMPer (10.19), characterized ER Ca2<sup>+</sup> uptake and release in cortical neurons and cerebellar Purkinje cells (Suzuki et al., 2014; Henderson et al., 2015). Interestingly, cerebellar Purkinje cells displayed differential ER Ca2<sup>+</sup> dynamics in postsynaptic compartment depending on the nature of synaptic inputs (Okubo et al., 2015). These lumenal ER Ca2<sup>+</sup> indicators also revealed interesting dynamics in dendritic spines, which suggest that [Ca2+]er and therefore [Ca2+]<sup>c</sup> can undergo synapse-specific regulation (Suzuki et al., 2014; Henderson et al., 2015).

### FUTURE PERSPECTIVES

Recent studies characterized the roles of presynaptic mitochondria and circuit-specific ER Ca2<sup>+</sup> mobility in dendrites directly via live imaging (Okubo et al., 2015; Kwon et al., 2016), however, organelle-specific Ca2<sup>+</sup> dynamics at local synapses

### REFERENCES


is only beginning to be explored. Genetically-encoded Ca2<sup>+</sup> sensors targeted to intracellular organelle and/or to specific synapses as well as functional indicators (like pHluorin-tagged synaptophysin or GluRs) will lead to the identification of synapse- and circuit-specific roles of mitochondria and ER Ca2<sup>+</sup> in neurons.

MCU has been recently shown to be associated with multiple regulatory proteins, which seems to modify or gate its gating properties and can prevent or enhance mitochondrial Ca2<sup>+</sup> uptake upon changes in cytosolic Ca2<sup>+</sup> dynamics (Perocchi et al., 2010; Mallilankaraman et al., 2012; Csordás et al., 2013; Plovanich et al., 2013; Raffaello et al., 2013; Sancak et al., 2013; De Stefani et al., 2015). In addition, MCU activity can be differentially controlled in different tissues (Fieni et al., 2012). Therefore, future investigations should probe the function of this MCU-regulatory complex in neurons and test if MCU and/or MCU-associated proteins can act as neuronal subtype-specific and/or synapse-specific functional modifiers.

In non-neuronal cells, ER and mitochondria establish focal connections which play a key role in Ca2<sup>+</sup> transfer from ER to mitochondria which has been characterized via intra- and inter-organelle Ca2<sup>+</sup> imaging (Rizzuto et al., 1993, 2012; Csordás et al., 2010; Kornmann, 2013). This transfer modulates ATP production in mitochondria and may also affect lipid exchange between these two organelles (Voelker, 1990; Cárdenas et al., 2010; Fujimoto and Hayashi, 2011). At present, in neurons, the role of Ca2<sup>+</sup> translocation between ER and mitochondria is largely unknown. Although immuno-EM images in vivo and Ca2<sup>+</sup> imaging with dyes in respiratory motor neurons suggested ER-mitochondria Ca2<sup>+</sup> crosstalk, future work will need to establish the context in which ER-mitochondria interface regulates Ca2<sup>+</sup> dynamics and synaptic function (Takei et al., 1992; Shoshan-Barmatz et al., 2004; Mironov and Symonchuk, 2006).

### AUTHOR CONTRIBUTIONS

All three authors co-wrote the manuscript.

### FUNDING

This work was partially funded by support from the NIH-NINDS (R01NS067557, FP), Japan society for the promotion of science fellowship for research abroad and The Uehara Memorial Foundation (YH), and a grant from the Human Frontier Science Program long-term fellowship (S-KK).


Neurosci. Biobehav. Rev. 37, 1211–1239. doi: 10.1016/j.neubiorev.2013. 04.011


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Kwon, Hirabayashi and Polleux. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Understanding Synaptogenesis and Functional Connectome in *C. elegans* by Imaging Technology

Jung-Hwa Hong1, 2 and Mikyoung Park 1, 3 \*

*<sup>1</sup> Center for Functional Connectomics, Korea Institute of Science and Technology, Seoul, South Korea, <sup>2</sup> Department of Life Sciences, Korea University, Seoul, South Korea, <sup>3</sup> Department of Neuroscience, Korea University of Science and Technology, Daejeon, South Korea*

Formation of functional synapses is a fundamental process for establishing neural circuits and ultimately for expressing complex behavior. Extensive research has interrogated how such functional synapses are formed and how synapse formation contributes to the generation of neural circuitry and behavior. The nervous system of *Caenorhabditis elegans*, due to its relatively simple structure, the transparent body, and tractable genetic system, has been adapted as an excellent model to investigate synapses and the functional connectome. Advances in imaging technology together with the improvement of genetically encoded molecular tools enabled us to visualize synapses and neural circuits of the animal model, which provide insights into our understanding of molecules and their signaling pathways that mediate synapse formation and neuronal network modulation. Here, we review synaptogenesis in active zones and the mapping of local connectome in *C. elegans* nervous system whose understandings have been extended by the advances in imaging technology along with the genetic molecular tools.

#### *Edited by:*

*George Augustine, Nanyang Technological University, Singapore*

#### *Reviewed by:*

*Fabrice Ango, University of Montpellier, France Antonio Malgaroli, Università Vita-Salute San Raffaele, Italy*

#### *\*Correspondence:*

*Mikyoung Park mikyoungpark7@gmail.com; mpark@kist.re.kr*

*Received: 25 April 2016 Accepted: 17 June 2016 Published: 29 June 2016*

#### *Citation:*

*Hong J-H and Park M (2016) Understanding Synaptogenesis and Functional Connectome in C. elegans by Imaging Technology. Front. Synaptic Neurosci. 8:18. doi: 10.3389/fnsyn.2016.00018* Keywords: presynaptic assembly, synaptic specificity, synaptogenesis, functional connectome, neural circuits, imaging, *C. elegans*

### INTRODUCTION

One of the fundamental goals of neuroscience is to understand the generation of functional nervous system that underlies neural basis of behavior and cognition. Extensive research has attempted to interrogate the molecular and cellular mechanisms of synapse formation and functional neural circuit development. Ever since it was proposed by Sydney Brenner in the mid 1960's (Brenner, 1974), the nematodes Caenorhabditis elegans (C. elegans) has been considered as an ideal model organism to study synaptic development and neural circuitry. The organism has relatively simple nervous system, having 302 neurons and its neurochemistry and genetics are similar to those of mammals. Moreover, the complete structure and connectivity of C. elegans nervous system have been deciphered through genetic screens and reconstruction of electron micrographs (EM) of serial sections, which led to discovery of novel molecules important for development and maintenance of functional synaptic connectivity (White et al., 1986). C. elegans with its transparent body was the first animal in which the green fluorescent protein (GFP) was expressed (Chalfie et al., 1994). Combined with its stable expression of fluorescently tagged proteins (Mello et al., 1991; Frokjaer-Jensen et al., 2008), studies with C. elegans have made major contributions to our knowledge on neural development, axonal migration, and synapse formation. Recently, selective plane illumination microscopy (SPIM) techniques such as tiling light-sheet SPIM (TLS-SPIM) (Fu et al., 2016) and inverted SPIM (iSPIM) (Wu et al., 2011) have been developed and utilized to achieve high spatiotemporal resolution 3-dimensional live imaging of C. elegans embyos with no detectable phototoxicity, which could enable studies on synaptogenesis and axon guidance during embryogenesis in C. elegans. Another recent work adopting complementation-activated light microscopy (CALM) in which proteins are conjugated with non-fluorescent split-fluorescent proteins, which become to be fluorescent when complemented with synthetic peptides enabled single-molecule imaging with a precision of 30 nm within synapses in live worms (Zhan et al., 2014).

Rapid developments of advanced imaging technologies have expanded our understanding of the molecular and cellular basis of synaptogenesis with great depth, taking a huge step closer to revealing functional neural connectome. Here, we discuss on the synaptogenesis in presynaptic active zones revealed by both conventional and advanced imaging set-ups and review recent work utilizing advanced imaging technology to unravel the functional connectome of neural circuits. Rather than dealing with the mechanistic aspects of synapse formation and neural circuits development, this review will mainly focus on how synaptic ultrastructure, synaptic formation, and functional neural connectome have been sophisticated by the advanced imaging technology. For more in-depth reviews on the mechanism of synaptogenesis, synaptic specificity, and neural circuits development, see Campbell et al. (2015), Cherra and Jin (2015), Jin (2015), Zhen and Samuel (2015), Yogev and Shen (2014), Chia et al. (2013), and Park and Shen (2012).

### IMAGING SYNAPSE ASSEMBLY

Chemical synapses are specialized intercellular junctions with two apposed compartments, the pre-synaptic terminal and the postsynaptic target, and the synaptic cleft which is about 20 nm gap between the pre- and postsynapses (Cowan and Kandel, 2001). Proper organization of pre- and postsynaptic components with precise regulation underlies formation of functional synapses. For the past decades, tremendous details regarding the morphology and assembly of C. elegans synaptic structure have been revealed with development of genetic tools and imaging technology. This section focuses on presynaptic assembly and synaptic specificity revealed by genetically encoded molecular tools and imaging technologies.

### Presynaptic Active Zone Imaging

The presynaptic compartment in C. elegans exhibits an overall structural organization similar to that in vertebrates, with synaptic vesicles clustered in and around the electron-dense membrane structure called active zone known to serve as a major site of neurotransmitter release. Ultrastructural analysis have shown that, despite the variations among the appearances, synapses of various organisms commonly display synaptic vesicle docking and fusion at active zone that can be identified by darkly stained membrane structures (Zhai and Bellen, 2004; Ackermann et al., 2015).

Many studies using C. elegans have investigated the role of various proteins localized at active zone in synapse formation (Yeh et al., 2005; Watanabe et al., 2011). Classical EM analysis has provided initial assessment of C. elegans synaptic components but its requirement for ultrathin sectioning of samples approximately 50 nm thickness (White et al., 1986) limits the resolution and impairs detailed visualization of fine structures. The multifunctional synaptic scaffolding protein SYD-2/liprin-α is one of the key proteins identified to regulate synaptic development in C. elegans and Drosophila (Zhen and Jin, 1999). The loss-of-function analysis on SYD-2/liprinα and uncoordinated-10 (UNC-10)/Rab3-interacting molecule (RIM), which is another dense-projection components (Weimer et al., 2006) revealed reduced vesicle recruitment at active zone (Stigloher et al., 2011; Kittelmann et al., 2013), and smaller denseprojection due to loss of SYD-2/liprin-α function (Kittelmann et al., 2013) unlike the finding showing an expanded denseprojection (Zhen and Jin, 1999). One suggested explanation for variability in syd-2 mutant synaptic ultrastructure is due to the differences in fixation procedure (Kittelmann et al., 2013). Nevertheless, it is certain that advanced and optimized imaging technique led to identification of regulatory proteins to retain synaptic vesicle at active zone.

A method which comprises of correlative fluorescence electron microscopy was developed and optimized to observe the nanoscopic localization of SYD-2/liprin-α in C. elegans active zone (Watanabe et al., 2011). The technique employed both stimulated emission depletion (STED) microscopy and photoactivated localization microscopy (PALM) on ultrathin sections for protein localization at super-resolution nanoscale level and subsequently correlate the protein localization with ultrastructures by electron microscope. The localization of SYD-2/liprin-α to the C. elegans presynaptic dense-projection observed by this technique (Watanabe et al., 2011) was consistent with the earlier finding from the immunoelectron micrograph (Yeh et al., 2005) but the result was more advanced to provide the precise localization of the proteins in small and dense structures likely within the synapse at the level of nanoscale super-resolution.

In addition, studies using advanced EM tomography of 250 nm thick sections combined with high-pressure freezing (HPF) and freeze substitution (Stigloher et al., 2011; Kittelmann et al., 2013) have resolved the highly complex structure of denseprojections at cholinergic neuromuscular junctions (NMJs) of C. elegans, revealing composition of building units forming baylike structures in which synaptic vesicles are docked to the active zone membrane. Furthermore, serial reconstruction of HPF EM sections and EM tomography enabled the construction of a high-resolution 3D model of presynaptic ultrastructure, overcoming resolution limitation raised by the conventional EM and revealing a physical link between dense-projections and synaptic vesicles within C. elegans presynaptic active zone.

### Presynaptic Assembly Imaging

Cell type-specific tagging of synaptic proteins with fluorescent reporter has been a key reagent to study synaptogenesis and its regulation in C. elegans (Nonet, 1999; Shen and Bargmann, 2003; Sieburth et al., 2005; Yeh et al., 2005). Hierarchical assembly of presynaptic active zone was observed in C. elegans HSNL synapses by fluorescently labeling the multiple active zone proteins and expressing them in the various mutant animals (Patel et al., 2006). Fluorescent protein fused with a synaptic vesicle-associated protein RAB-3 visualized synaptic vesicle clusters and confirmed the presynaptic localization of various active zone components, including SYD-1, SYD-2/liprin-α, ELKS-1/ERC/CAZ-associated structural protein (CAST), GIT, and SAD-1 kinase in the HSNL synapses (Patel et al., 2006). Altering the location of SYG-1/Neph1 by ectopically expressing SYG-2/Nephrin in the secondary vulval epithelial cells, led to ectopic localizations of presynaptic components, including RAB-3, SYD-1, SYD-2/liprin-α, GIT, and ELKS-1/ERC/CAST in the HSNL regions where the secondary vulval epithelial cells made contacts (**Figure 1A**). This supported the idea that along with SYG-2/Nephrin as an upstream signal of SYG-1/Neph1, SYG-1/Neph1 defines the presynaptic localization and is sufficient to recruit presynaptic components, including the two key scaffold molecules SYD-1 and SYD-2/liprin-α (Zhen and Jin, 1999) to the regions defined by its localization (Patel et al., 2006; **Figure 1B**).

In syd-1 and syd-2 mutants, the presynaptic components, including RAB-3, ELKS-1/ERC/CAST, GIT, SAD-1,

FIGURE 1 | Synaptic specificity regulated by non-neuronal factors. (A) Synaptic connectivity of neurons and muscles associated in the egg-laying circuit of *C. elegans*. HSNL forms synapses with vulval muscle 2 (vm2) and ventral cord (VC) motor neurons, VC4 and VC5 specifically to the regions immediately adjacent to the primary epithelial cells (1◦ ) which secretes SYG-2/Nephrin. Mutations in SYG-1/Neph1 or SYG-2/Nephrin disrupt synaptic specificity of HSNL and cause ectopic synapse formation with select body wall muscle (BWM). Ectopic positioning of SYG-2/Nephrin to the secondary epithelial cells (2◦ ) recruited SYG-1/Neph1 to HSNL near the secondary epithelial cells, which was shown to be sufficient to form synapses ectopically at the sites where SYG-1/Neph1 is recruited (yellow circles). (B) Pathway for HSNL synapse assembly. SYG-2/Nephrin ensures proper localization of SYG-1/Neph1 which defines presynaptic localization of the active zone proteins. ELKS-1/ERC/CAST could function redundantly with SYD-1 or other unidentified presynaptic proteins that positively regulate synapse assembly (gray lines). In the presence of SYD-1, the SYD-2/liprin-α and ELKS-1/ERC/CAST interaction was enhanced (red arrow). In the presence of RSY-1, SYD-1, and ELKS-1/ERC/CAST interaction is weakened (solid green), suggesting RSY-1 as a negative regulator in the HSNL presynaptic assembly process likely by weakening the SYD-2/liprin-α and ELKS-1/ERC/CAST interaction (dotted green) indirectly through the RSY-1 and SYD-1 interaction. Plain lines indicate biochemical interactions. (C) Synaptic connectivity of AIY and RIA interneurons regulated by ventral cephalic sheath cells (CEPshV) at *C. elegans* nerve ring. Synapses between AIY and RIA are formed *en passent* as they are ensheathed in zone 2 by CEPshV, which secretes UNC-6/Netrin that regulates UNC-40/DCC activity in AIY. Abnormal distend positioning of CEPshV toward zone 1 causes ectopic localizations of both presynapses (red circles) and UNC-40/DCC (purple triangles) in zone 1 of AIY. (D) Pathways for AIY and RIA connectivity. CEPshV secretes UNC-6/Netrin, which regulates both positioning of presynapses in AIY and axon guidance of postsynaptic RIA through UNC-40/DCC activity to the location specified by CEPshV.

UNC-57/endophilin, and SNN-1/synapsin-1 were failed to be assembled, identifying those presynaptic components as downstream molecules of SYD-1 and SYD-2/liprin-α in the active zone assembly process (Patel et al., 2006). Gain-of-function mutation (Dai et al., 2006) or overexpression of SYD-2/liprin-α (Patel et al., 2006) in syd-1 mutants completely restored the synaptic accumulation of SNB-1/synaptobrevin, whereas the SYD-1 overexpression in syd-2 mutants was not sufficient to induce the rescue effect (Patel et al., 2006), illustrating the SYD-1 and SYD-2/liprin-α mediated presynaptic assembly with the SYD-1 as an upstream of SYD-2/liprin-α (**Figure 1B**). Although the loss of ELKS-1 function by itself did not induce apparent defects in synapse assembly in C. elegans HSNL synapses (Dai et al., 2006; Patel et al., 2006), synapse formation in the syd-2 gain-of-function and syd-1 double mutants exhibited a high dependency on ELKS-1 expression (Dai et al., 2006), suggesting that ELKS-1 functions redundantly with SYD-1 or other presynaptic proteins that positively regulate synapse assembly (**Figure 1B**).

Regulator of synaptogenesis-1 (RSY-1) was cloned as a negative regulator of synapse formation for its deletion mutants to lead to extra synapse formation and exhibit increased accumulation of SNB-1/synaptobrevin at presynaptic sites in the HSNL (Patel and Shen, 2009). A single-cell in situ protein-protein interaction assay revealed enhanced interaction between SYD-2/liprin-α and ELKS-1/ERC/CAST in presence of SYD-1 while direct interaction between SYD-1 and ELKS-1/ERC/CAST is weakened in presence of RSY-1, suggesting RSY-1 as a negative regulator of C. elegans HSNL synapse assembly likely by weakening the SYD-2/liprin-α and ELKS-1/ERC/CAST interaction indirectly through its interaction with SYD-1 (**Figure 1B**). Together, presynaptic differentiation at C. elegans HSNL synapses was initiated by SYG-1/Neph1, a synaptic specificity molecule that defines the location of presynaptic sites along the HSNL axon, leading to activate the presynaptic assembly process by recruiting the two key scaffolding proteins SYD-1 and SYD-2/liprin-α. SYD-2/liprinα-centered assembly of presynaptic components was achieved through the inter-communications among positive (SYD-1 and ELKS-1/ERC/CAST) and negative (RSY-1) regulators (**Figure 1B**).

### IMAGING SYNAPTIC SPECIFICITY

During the event of synapse formation, a precise apposition between the presynaptic release sites and postsynaptic receptors must be accomplished to ensure a rapid neurotransmitter release and reliable synaptic response. Neurons can select subpopulations of neurons they form synapses onto and can also select the defined specific subcellular sites to establish synapses (Akins and Biederer, 2006; White, 2007; Margeta and Shen, 2010). Such synaptic specificity is achieved by trans-synaptic adhesion between pre- and postsynaptic neurons (Yamagata et al., 2002; Graf et al., 2004; Choe et al., 2006), adhesion between the presynaptic neuron and a guidepost cell (Shen and Bargmann, 2003; Shen et al., 2004), molecules secreted from pre- or postsynaptic neurons (Umemori et al., 2004; Inaki et al., 2007) or from a guidepost cell (Christopherson et al., 2005; Colon-Ramos et al., 2007). In C. elegans, synaptic contacts are usually formed en passant, in which synapses are formed along the adjacent processes but not its terminus (White et al., 1986). Synaptic specificity studies in C. elegans have been accelerated upon the development of the expression tool of fluorescently tagged proteins in specific cell types driven by cell type-specific promoters, which enabled researchers to specifically label pre-, postsynapses, and neighboring guidepost cells (Nonet, 1999; Shen and Bargmann, 2003; Grunwald et al., 2004; Francis et al., 2005; Sieburth et al., 2005; Yeh et al., 2005; Hoerndli et al., 2013).

A specific synaptic connectivity between amphid interneuron Y (AIY) and ring interneuron A (RIA) in C. elegans nerve ring, considered as brain of the animal, was fluorescently visualized by expressing presynaptic RAB-3 in AIY and postsynaptic glutamate receptors GLR-1 in RIA (Colon-Ramos et al., 2007; Shao et al., 2013) (**Figure 1C**). The localization of synaptic connectivity between AIY and RIA has shown to be restricted in the zone 2 of AIY axon (**Figure 1C**) and such specificity is achieved by activation of both UNC-6/Netrin, a well-known axon guidance molecule that is exclusively expressed by glialike ventral cephalic sheath cells (CEPshV) (Wadsworth et al., 1996) and the netrin receptor UNC-40/Deleted in Colorectal Cancer (DCC) (Colon-Ramos et al., 2007), supporting the idea that secreted molecules from glia govern synaptic specificity. Confocal microscopy revealed the projection of the CEPshV processes with respect to the region of innervation between AIY and RIA (**Figure 1C**). Loss-of-function in either UNC-34/enabled, a regulator of the actin cytoskeleton (Colon-Ramos et al., 2007) or circuit maintenance abnormal protein (CIMA-1), a regulator of synaptic maintenance in C. elegans (Shao et al., 2013), caused morphological alterations in CEPshV which migrated toward further posteriorly to ensheath AIY axon in zone 1 (**Figure 1C**). Morphological alterations in CEPshV led to ectopic localization of both UNC-40/DCC and presynaptic components in zone 1 (**Figure 1C**) due to the existence of UNC-6/Netrin secreted from CEPshV in zone 1 (Colon-Ramos et al., 2007). The process of RIA in unc-34 mutants also abnormally migrated toward zone 1 where the ectopic synapses were formed (**Figure 1C**). Together, UNC-40/DCC plays two independent roles in each neuron, which are positioning of presynapses in AIY and axon guidance of postsynaptic RIA to the location specified by CEPshV (**Figure 1D**). These findings further support the model of non-neuronal contribution to the regulation of precise localization of synaptogenesis.

Earlier than the AIY-RIA synaptic specificity study, the C. elegans egg-laying circuit, which is predominantly innervated by the two hermaphrodite-specific motor neurons (HSNs), HSNL and HSNR, and the two ventral cord (VC) motor neurons, VC4 and VC5 has been reported to be regulated by non-neuronal factor. HSNL and HSNR synapse onto vulval muscle cells and onto the VC4 and VC5 neurons, while VC4 and VC5 neurons also synapse onto the vulval muscle cells. Despite the direct contact between HSN and VC processes, synapses formed between these cells are only restricted to the regions adjacent to the vulva (White et al., 1986) (**Figure 1A**). The specific positioning of synapses and the recognition between HSNL and its target were determined by adjacent vulva epithelial guidepost cells that express SYG-2/Nephrin. SYG-2/Nephrin interacts with SYG-1/Neph1 expressed in the HSNL, to recruit SYG-1/Neph1 to the site along the HSNL axon where presynaptic sites are developed (Shen and Bargmann, 2003; Shen et al., 2004) (**Figure 1A**).

More recently, introduction of the GFP reconstitution across synaptic partners (GRASP) developed in C. elegans has overcome the challenges addressed by labor-intensive conventional EM analysis and increased the spatial resolution to visualize the pre- and postsynaptic contacts. GRASP is based on functional complementation between two non-fluorescent split-GFP fragments separately expressed in the pre- and postsynaptic neurons, which label synapses between two cells of close proximity in living animals (Feinberg et al., 2008). Using GRASP, specific visualization of synaptic contacts between AIY and RIA was observed with high spatial resolution (Shao et al., 2013). In addition, GRASP revealed restricted synaptic localization between AIY and CEPshV (Shao et al., 2013), which is consistent with the published EM data (White et al., 1986). Formation of ectopic synapses between AIY and CEPshV due to morphological alteration in CEPshV was confirmed as well (Shao et al., 2013) (**Figure 1C**). GRASP application has also confirmed the SYG-1/Neph1 and SYG-2/Nephrin as synaptic specificity regulators of HSN synapses with vulval muscles and VC neurons. Analyzing GRASP fluorescence in wild-type and syg-1 or syg-2 mutants recapitulated the synaptic connectivity of HSN neurons (Feinberg et al., 2008) (**Figure 1A**). Besides the C. elegans nervous system, the GRASP has also been widely adapted by other model systems, such as Drosophila (Gordon and Scott, 2009; Gong et al., 2010) mouse (Kim et al., 2012; Yamagata and Sanes, 2012) and the cultured hippocampal neuronal system (Tsetsenis et al., 2014). Lately, newly modified GRASP strategies, involving activity-dependent synaptic GRASP and multi-color fluorescence reconstitution across synapses (X-RASP) have been validated in Drosophila, allowing preferential labeling of active synapses and multi-color labeling of active synapses in one animal (Macpherson et al., 2015; Li et al., 2016). Continuous development of GRASP shows the potential to expand the utility of GRASP to identify and map synaptic connectivity of neural circuits in the living animal with high resolution.

### IMAGING FUNCTIONAL NEURAL CIRCUITS

An underlying goal of neuroscience is to understand the neural connectome that are responsible for synaptic function and neuronal basis of behavior. Anatomical structural connectome of the whole nervous system of C. elegans, which has been fully mapped by EM of serial sections (White et al., 1986), has served as a useful resource for researchers to study circuit function, thus making the C. elegans nervous system as an excellent model to investigate functional connectome of neural circuits. For the past decade, optogenetics has been widely adapted to manipulate neural circuits and examine the corresponding changes in synaptic function and behavior (Fang-Yen et al., 2012; Husson et al., 2013). Optogenetics uses genetically encoded light-sensitive proteins such as channelrhodopsins (Nagel et al., 2003, 2005), halorhodopsins (Han and Boyden, 2007; Zhang et al., 2007; Husson et al., 2012b), and archaerhodopsins (Ihara et al., 1999) as optogenetic actuators to either activate or inhibit neuronal activity via light and genetically encoded sensors such as GCaMP calcium indicator (Tian et al., 2009) and Clomeleon chloride indicator (Kuner and Augustine, 2000; Berglund et al., 2006) as optogenetic sensors to monitor responses to the synaptic inputs. This section will discuss various experimental imaging approaches to interrogate the neural connection using the C. elegans nervous system.

The initial optogenetics was applied to manipulate the behavior of C. elegans (Nagel et al., 2005). Expression of Channelrhodopsin-2 (ChR2), a blue light-gated depolarizing cation channel used to activate neural activity in C. elegans body muscles caused blue light-evoked contractions (Nagel et al., 2003, 2005), whereas expression of NpHR, a yellow light-gated hyperpolarizing chloride ion pump applied to inhibit neural activity in C. elegans muscle cells caused an extension of the worm's body length and locomotion defects by whole-field illumination of yellow-green light (Zhang et al., 2007; Husson et al., 2012b). Aside from NpHR, a yellow-green light-sensitive archaerhodopsin-3 (Arch) (Ihara et al., 1999) and a greenblue light-sensitive Mac (Waschuk et al., 2005) have also been expressed in C. elegans and induced a stronger optical silencing effect than NpHR likely due to efficient protein trafficking to the plasma membrane (Chow et al., 2010; Husson et al., 2012b). The simultaneous use of Arch and Mac enabled inhibition of two different neuronal subpopulations, depending on the illuminating lights used.

Light-sensitive probes expressed in C. elegans in vivo are mostly under the control of a promoter sequence. However, promoter-driven single cell expression of optogenetic protein is challenging to achieve due to the lack of single cell-specific promoter and instead proteins are diversely expressed, eliciting robust behavioral responses upon whole-field illumination (Husson et al., 2013). Although it may be useful for inspecting a novel optogenetic protein, optical manipulation of individual neurons needs to be accomplished in order to obtain insights into individual contribution by single neurons in functional connectivity. To this aim, new methods have been adapted in C. elegans to drive selective optical manipulation, either by genetically modulated single cell-specific expression of optogenetic protein (Ezcurra et al., 2011; Schmitt et al., 2012; Cho and Sternberg, 2014; Guo et al., 2015) or selective illumination of target neurons with a high spatial and temporal resolution (Guo et al., 2009; Leifer et al., 2011; Stirman et al., 2011; Husson et al., 2012a,b; Kocabas et al., 2012; Cohen et al., 2014; Luo et al., 2014; Shipley et al., 2014; Trojanowski et al., 2014) in order to dissect functional connections within the neural circuits (**Table 1**).

Mainly adapted approach to specifically deliver light-sensitive opsins to individual neurons of C. elegans restricts the opsin expression by genetic application using Cre or FLP recombinases (Ezcurra et al., 2011; Schmitt et al., 2012; Cho and Sternberg, 2014; Guo et al., 2015) (**Figure 2**). The recombinase-dependent gene expression is driven by a set of two promoters, a first promoter driving the expression of opsin conjugated with a


#### TABLE 1 | Cell-specific optogenetic applications in *C. elegans*

\**Optogenetic manipulation driven by neuronal type-specific promoters rather than Cre/FLP recombinase application.*

fluorophore along with or without a bicistronic fluorescent reporter and a second promoter driving the expression of Cre or FLP recombinase. In the first promoter-containing construct, a transcription termination sequence flanked by recombinase recognition sequences, loxP or FRT that are recognized by Cre or FLP recombinase is enclosed in front of opsin. The Cre or FLP recombinase-mediated recombination of loxP or FRT sites excised the stop sequence and allows conditional expression of opsin only in the target cell where both promoters are active (Husson et al., 2013) (**Figure 2**). Using Cre and FLP system, ChR2 were specifically expressed in PVC interneurons which evoked forward locomotion and in AVA interneuron and ASH sensory neurons which evoked backward-movement upon photostimulation (Ezcurra et al., 2011; Schmitt et al., 2012) (**Table 1**). Further effort to isolate exclusive expression of the light-sensitive proteins in a single cell (Ezcurra et al., 2011) would need to define the role of individual single neurons in functional neural circuits.

Instead of using genetically generated system and whole-field illumination, spatiotemporally patterned illumination of neurons expressing light-sensitive optogenetic proteins in immobilized C. elegans was used by for the first time in vivo using a digital micromirror device (DMD) whose individual mirrors can be controlled independently to precisely determine the location and size of the regions to be illuminated while simultaneously recording the calcium levels using a genetically encoded calcium sensor, GCaMP to analyze the functional connections among neurons. Combining the optogenetic actuator ChR2 and the sensor GCaMP with the patterned illumination via a DMD technology, the functional connections from the sensory neuron ASH to the interneurons AVA and AVD and the connections between the interneurons RIM and AVA have been monitored (Guo et al., 2009; **Table 1**).

Improvement in microscopic analysis and optogenetic illumination system allowed manipulation of neural activity in a freely behaving C. elegans with a high spatiotemporal resolution, providing an in-depth analysis on functional neural circuits underlying behavior at a single-cell level. A modified threepanel liquid crystal display (3-LCD) projector for simultaneous multicolor illumination and a motorized X-Y stage for keeping the unrestrained worm centered in the camera's field of view with a standard inverted epifluorescence microscope were systemized (Stirman et al., 2011; Husson et al., 2012b) and the Colbert system was equipped to control locomotion and behavior in real time (Leifer et al., 2011; Luo et al., 2014; Shipley et al., 2014). Spatial regulation of optical illumination is controlled either by estimating the coordinates of targeted cells using the machine-vision algorithms (Leifer et al., 2011; Trojanowski et al., 2014) or by calculating the anterior-posterior (A-P) axis (Stirman et al., 2011). Both systems have been instrumental in defining neural coding of several behaviors in C. elegans linked to the motor circuit, avoidance circuit, nociceptive circuit, chemotaxis circuit, and feeding circuits of freely moving worms (**Table 1**). Using AIY expressing ChR2 and targeted illumination by the DMD technology, it was shown that optogenetic manipulation of AIY activity alone was sufficient to evoke chemotactic behavior in freely moving C. elegans, and was suggested that AIY is plausible to act as a control node for coordinating other taxis behaviors as well (Kocabas et al., 2012). Another report using the Colbert system equipped with

the DMD investigated an experience-dependent salt chemotaxis circuit. Optogenetic manipulation of neuronal activity of the ASER sensory neuron expressing ChR2 was shown to be connected to positive and negative chemotaxis in response to salt concentrations, indicating that ASER sensory neuron encodes the perception of salt concentration and the memory of the chemotactic set point in a chemotaxis circuit of C. elegans (Luo et al., 2014). In addition, optogenetic manipulations of specific pharyngeal neurons MC, M2, M4, and I1 in freely behaving worms by adopting ChR2 for optical stimulation and Mac for optical silencing along with the DMD for targeted illumination revealed a pharyngeal pumping/feeding circuit and identified the regulation of feeding rate by nicotinic and muscarinic receptors through the pharyngeal neuronal network (Trojanowski et al., 2014). Furthermore, multispectral illumination (Stirman et al., 2011) enables simultaneous application of optical stimulation and inhibition to an individual animal. Emerging studies have successfully facilitated multimodal optogenetic manipulation on C. elegans to independently excite different neurons in a single worm (Erbguth et al., 2012; Husson et al., 2012b; Schild and Glauser, 2015).

### PERSPECTIVES

C. elegans is currently the best organism to study synapses and neuronal circuits because the connectivity of its 302 neurons has been well-defined by serial reconstruction of EM (White et al., 1986), the body is transparent, and it is a genetically tractable animal model. C. elegans was one of the first organisms that GFP was expressed to label protein (Chalfie et al., 1994), GRASP was utilized to visualize specific synaptic contacts (Feinberg et al., 2008), optogenetics was applied to manipulate behavior of live animals (Nagel et al., 2005), and more recently, sonogenetics using low-pressure ultrasound was challenged to activate specific ultrasonically sensitized neurons and modify locomotory behavior (Ibsen et al., 2015).

As genetically encoded fluorescent proteins have been rapidly developed for the past decades since the GFP was introduced in the field, it is also expected that the number of optogenetic tools will rapidly increase to likely provide optogenetic proteins with different spectral properties (Zhang et al., 2008; Gradinaru et al., 2010) and ionic specificities (Han and Boyden, 2007; Zhang et al., 2007) and help expand the understanding of synaptic function and neural circuits. During such processes, it is confidently predicted that C. elegans will provide a systematic in vivo platform to test the optogenetic tools newly developed and to ultimately apply to the synaptic function and functional connectome studies. Together with the improvement of fluorescent and optogenetic tools, continuous development in C. elegans imaging technology will promise a breakthrough in deciphering functional neural connectome.

In addition to the monitoring and controlling of existing neuronal circuits via optogenetic applications and advanced microscopy systems as described in this review, it is very plausible to develop the ways to actively manipulate neural circuits for instance by inserting new connections or removing existing connections, resulting in the reprogramming of neural circuits. Indeed, a recent study on artificial modifications of neural circuits was reported in C. elegans by expressing transgenically targeted heterologous connexin to insert a new electrical synapse between normally unconnected neurons in intact animals, which resulted in altered salt taste and olfactory chemotaxis behavior (Rabinowitch et al., 2014). Conversely, laser ablation method can be used to remove existing connection (Sulston and White, 1980; Bargmann et al., 1993; McIntire et al., 1993; Fang-Yen et al., 2012; Rabinowitch et al., 2013). Such artificial modification of neural circuits not only help understand fundamental functions of neuronal connectivity underlying complex behavior but could also be applied to disease brain circuits with the purpose of therapeutics at the circuit level.

### AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

### REFERENCES


### ACKNOWLEDGMENTS

The work in M. Park laboratory was supported by the KIST Institutional Programs (Project No. 2E26190 and 2E26170).


identity lineaging and neurodevelopmental imaging in Caenorhabditis elegans. Proc. Natl. Acad. Sci. U.S.A. 108, 17708–17713. doi: 10.1073/pnas.1108494108


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Hong and Park. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.